<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/">
    <channel>
        <title>Jake Brukhman</title>
        <link>https://paragraph.com/@brukhman</link>
        <description>Founder &amp; CEO @ CoinFund</description>
        <lastBuildDate>Mon, 16 Nov 2026 00:24:23 GMT</lastBuildDate>
        <docs>https://validator.w3.org/feed/docs/rss2.html</docs>
        <generator>https://github.com/jpmonette/feed</generator>
        <language>en</language>
        <image>
            <title>Jake Brukhman</title>
            <url>https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/62a1efcd5cbf8d97254219b4039bf4c0dc81a9d207e3257492d994c49e76b975.jpg</url>
            <link>https://paragraph.com/@brukhman</link>
        </image>
        <copyright>All rights reserved</copyright>
        <item>
            <title><![CDATA[Using Jargon CLI in Jargon v0.2.0]]></title>
            <link>https://paragraph.com/@brukhman/using-jargon-cli-in-jargon-v0-2-0</link>
            <guid>Rg4uTvJ98dsMN1t3qLYc</guid>
            <pubDate>Mon, 08 May 2023 03:07:42 GMT</pubDate>
            <description><![CDATA[The Jargon CLI is a new feature of Jargon: it is a command line shell that works with the OpenAI API and allows you to experiment with Jargon programs quickly. You can try it out in v0.2.0 at the Jargon Github repoInstallationTo install it, simply use pip:$ cd gpt-jargon $ pip install . For the CLI to be able to access your LLM, make sure your OPENAI_API_KEY environment variable is set. The default model used is gpt-4, but you may want to use an alternative model with the --model option (or t...]]></description>
            <content:encoded><![CDATA[<p>The Jargon CLI is a new feature of Jargon: it is a command line shell that works with the OpenAI API and allows you to experiment with Jargon programs quickly. You can try it out in <code>v0.2.0</code> at the <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://github.com/jbrukh/gpt-jargon">Jargon Github repo</a></p><h1 id="h-installation" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Installation</h1><p>To install it, simply use <code>pip</code>:</p><pre data-type="codeBlock" text="$ cd gpt-jargon
$ pip install .
"><code><span class="hljs-meta prompt_">$ </span><span class="bash"><span class="hljs-built_in">cd</span> gpt-jargon</span>
<span class="hljs-meta prompt_">$ </span><span class="bash">pip install .</span>
</code></pre><p>For the CLI to be able to access your LLM, make sure your <code>OPENAI_API_KEY</code> environment variable is set. The default model used is <code>gpt-4</code>, but you may want to use an alternative model with the <code>--model</code> option (or the <code>JARGON_CLI_MODEL</code> environment variable).</p><h1 id="h-using-the-cli" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Using the CLI</h1><p>First, turn on your CLI:</p><pre data-type="codeBlock" text="$ jargon cli
Commands: /exit, /ls, /cat &lt;proc&gt;, /edit &lt;proc&gt;, /execute &lt;proc&gt; or /&lt;proc&gt;, /clear
user&gt; 
"><code>$ jargon cli
Commands: <span class="hljs-operator">/</span>exit, <span class="hljs-operator">/</span>ls, <span class="hljs-operator">/</span>cat <span class="hljs-operator">&#x3C;</span>proc<span class="hljs-operator">></span>, <span class="hljs-operator">/</span>edit <span class="hljs-operator">&#x3C;</span>proc<span class="hljs-operator">></span>, <span class="hljs-operator">/</span>execute <span class="hljs-operator">&#x3C;</span>proc<span class="hljs-operator">></span> or <span class="hljs-operator">/</span><span class="hljs-operator">&#x3C;</span>proc<span class="hljs-operator">></span>, <span class="hljs-operator">/</span>clear
user<span class="hljs-operator">></span> 
</code></pre><p>(Some of these commands you can also run right from your bash shell. See <code>jargon --help</code> for details.)</p><p>The first thing to do is to edit a Jargon procedure so you can run it. We do this using the <code>/edit</code> command:</p><h3 id="h-editing-jargon-procedures" class="text-2xl font-header !mt-6 !mb-4 first:!mt-0 first:!mb-0">Editing Jargon procedures</h3><pre data-type="codeBlock" text="user&gt; /edit cats
"><code><span class="hljs-meta prompt_">user> </span><span class="bash">/edit cats</span>
</code></pre><p>Alternatively, in bash:</p><pre data-type="codeBlock" text="bash$ jargon edit cats
"><code>bash<span class="hljs-variable">$ </span>jargon edit cats
</code></pre><p>This opens an editor which edits the file <code>~/.jargon/cats.jarg</code>. You can see all your Jargon procedure using <code>/ls</code>:</p><pre data-type="codeBlock" text="user&gt; /ls
* cats.jarg
"><code>user<span class="hljs-operator">></span> <span class="hljs-operator">/</span>ls
<span class="hljs-operator">*</span> cats.jarg
</code></pre><p>Or, in bash:</p><pre data-type="codeBlock" text="bash$ jargon ls
* cats.jarg
"><code>bash$ jargon ls
<span class="hljs-operator">*</span> cats.jarg
</code></pre><p>To quickly list the code of a procedure, you can use <code>/cat</code>:</p><pre data-type="codeBlock" text="user&gt; /cat cats
+++ cats
/output [a random name for a cat]
+++
"><code>user<span class="hljs-operator">></span> <span class="hljs-operator">/</span>cat cats
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> cats
<span class="hljs-operator">/</span>output [a random name <span class="hljs-keyword">for</span> a cat]
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><h3 id="h-executing-jargon-procedures" class="text-2xl font-header !mt-6 !mb-4 first:!mt-0 first:!mb-0">Executing Jargon procedures</h3><p>All of the following commands do the same thing, namely execute<code>cats.jarg</code>:</p><pre data-type="codeBlock" text="user&gt; /execute cats
user&gt; /execute cats.jarg
user&gt; /cats
user&gt; /cats.jarg
"><code>user<span class="hljs-operator">></span> <span class="hljs-operator">/</span>execute cats
user<span class="hljs-operator">></span> <span class="hljs-operator">/</span>execute cats.jarg
user<span class="hljs-operator">></span> <span class="hljs-operator">/</span>cats
user<span class="hljs-operator">></span> <span class="hljs-operator">/</span>cats.jarg
</code></pre><p>Or, in bash:</p><pre data-type="codeBlock" text="bash$ jargon execute cats
"><code>bash$ jargon <span class="hljs-built_in">execute</span> cats
</code></pre><p>Here it is in action. Note that while <em>commands</em> to the CLI start with a <code>/</code>, normal input is interpreted intelligently by the LLM:</p><pre data-type="codeBlock" text="user&gt; /cats
+++ cats
/output [a random name for a cat]
+++
jargon&gt; Output: Luna
user&gt; /cats
+++ cats
/output [a random name for a cat]
+++
jargon&gt; Output: Simba
user&gt; run it again
jargon&gt; Output: Oliver
user&gt; again
jargon&gt; Output: Bella
user&gt;
"><code>user<span class="hljs-operator">></span> <span class="hljs-operator">/</span>cats
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> cats
<span class="hljs-operator">/</span>output [a random name <span class="hljs-keyword">for</span> a cat]
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
jargon<span class="hljs-operator">></span> Output: Luna
user<span class="hljs-operator">></span> <span class="hljs-operator">/</span>cats
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> cats
<span class="hljs-operator">/</span>output [a random name <span class="hljs-keyword">for</span> a cat]
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
jargon<span class="hljs-operator">></span> Output: Simba
user<span class="hljs-operator">></span> run it again
jargon<span class="hljs-operator">></span> Output: Oliver
user<span class="hljs-operator">></span> again
jargon<span class="hljs-operator">></span> Output: Bella
user<span class="hljs-operator">></span>
</code></pre><h3 id="h-clearing-the-memory-exiting" class="text-2xl font-header !mt-6 !mb-4 first:!mt-0 first:!mb-0">Clearing the memory, exiting</h3><p>We can wipe all memory of our conversation:</p><pre data-type="codeBlock" text="user&gt; /clear
"><code><span class="hljs-meta prompt_">user> </span><span class="bash">/clear</span>
</code></pre><p>We can also exit the CLI:</p><pre data-type="codeBlock" text="user&gt; /exit
"><code><span class="hljs-meta prompt_">user> </span><span class="bash">/exit</span>
</code></pre><h2 id="h-some-fun-jargon-procedures-to-try" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Some fun Jargon procedures to try</h2><h3 id="h-idea-evaluation" class="text-2xl font-header !mt-6 !mb-4 first:!mt-0 first:!mb-0">Idea evaluation</h3><pre data-type="codeBlock" text="+++ idea-score
@start:
  /output &quot;Enter your idea:&quot;
  /wait $idea
  -- Evaluate the $idea and let $score = [the score of the $idea from 0 to 100]
  /output &quot;[up to three word summary of $idea]: $score% ([very short summary of why you gave this score])&quot;
  /goto @start
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> idea<span class="hljs-operator">-</span>score
@start:
  <span class="hljs-operator">/</span>output <span class="hljs-string">"Enter your idea:"</span>
  <span class="hljs-operator">/</span>wait $idea
  <span class="hljs-operator">-</span><span class="hljs-operator">-</span> Evaluate the $idea and let $score <span class="hljs-operator">=</span> [the score of the $idea <span class="hljs-keyword">from</span> <span class="hljs-number">0</span> to <span class="hljs-number">100</span>]
  <span class="hljs-operator">/</span>output <span class="hljs-string">"[up to three word summary of $idea]: $score% ([very short summary of why you gave this score])"</span>
  <span class="hljs-operator">/</span>goto @start
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><p>And some of the output from <code>jargon execute idea-score</code>:</p><pre data-type="codeBlock" text="jargon&gt; Enter your idea:
user&gt; A dating service for fish.
jargon&gt; Aquatic romance platform: 34% (Unusual niche, limited practicality)
user&gt; An opaque fish tank.
jargon&gt; Invisible aquatic habitat: 12% (Poor visibility, defeats purpose)
user&gt; A superintelligent LLM.
jargon&gt; Advanced AI thinker: 89% (High potential, ethical considerations)
"><code>jargon<span class="hljs-operator">></span> Enter your idea:
user<span class="hljs-operator">></span> A dating service <span class="hljs-keyword">for</span> fish.
jargon> Aquatic romance platform: <span class="hljs-number">34</span><span class="hljs-operator">%</span> (Unusual niche, limited practicality)
user<span class="hljs-operator">></span> An opaque fish tank.
jargon> Invisible aquatic habitat: <span class="hljs-number">12</span><span class="hljs-operator">%</span> (Poor visibility, defeats purpose)
user<span class="hljs-operator">></span> A superintelligent LLM.
jargon> Advanced AI thinker: <span class="hljs-number">89</span><span class="hljs-operator">%</span> (High potential, ethical considerations)
</code></pre><h3 id="h-rick-and-morty-text-adventure-game" class="text-2xl font-header !mt-6 !mb-4 first:!mt-0 first:!mb-0">Rick and Morty text adventure game</h3><pre data-type="codeBlock" text="+++ rick-and-morty
* The series Rick and Morty primarily follows the adventures of the eccentric and alcoholic scientist Rick Sanchez and his good-hearted but easily influenced grandson Morty Smith. They embark on dangerous and often absurd adventures across the multiverse, using Rick&apos;s portal gun to travel between different dimensions and alternate realities. The show combines elements of science fiction, dark humor, and philosophical concepts, which serve as a backdrop for exploring the characters&apos; personal lives and relationships.
* You are Morty.
@loop:
  -- $universe = [pick a crazy and random Rick and Morty universe]
  -- $story = [generate a story based on a $universe]
  -- Describe the scene briefly
  -- Give me 4 options of how I may proceed from here.
  /wait $option
  -- $topic = $option
  /goto @loop
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> rick<span class="hljs-operator">-</span>and<span class="hljs-operator">-</span>morty
<span class="hljs-operator">*</span> The series Rick and Morty primarily follows the adventures of the eccentric and alcoholic scientist Rick Sanchez and his good<span class="hljs-operator">-</span>hearted but easily influenced grandson Morty Smith. They embark on dangerous and often absurd adventures across the multiverse, <span class="hljs-keyword">using</span> <span class="hljs-title">Rick</span>'<span class="hljs-title">s</span> <span class="hljs-title">portal</span> <span class="hljs-title">gun</span> <span class="hljs-title">to</span> <span class="hljs-title">travel</span> <span class="hljs-title">between</span> <span class="hljs-title">different</span> <span class="hljs-title">dimensions</span> <span class="hljs-title">and</span> <span class="hljs-title">alternate</span> <span class="hljs-title">realities</span>. <span class="hljs-title">The</span> <span class="hljs-title">show</span> <span class="hljs-title">combines</span> <span class="hljs-title">elements</span> <span class="hljs-title">of</span> <span class="hljs-title">science</span> <span class="hljs-title">fiction</span>, <span class="hljs-title">dark</span> <span class="hljs-title">humor</span>, <span class="hljs-title">and</span> <span class="hljs-title">philosophical</span> <span class="hljs-title">concepts</span>, <span class="hljs-title">which</span> <span class="hljs-title">serve</span> <span class="hljs-title"><span class="hljs-keyword">as</span></span> <span class="hljs-title">a</span> <span class="hljs-title">backdrop</span> <span class="hljs-title"><span class="hljs-keyword">for</span></span> <span class="hljs-title">exploring</span> <span class="hljs-title">the</span> <span class="hljs-title">characters</span>' <span class="hljs-title">personal</span> <span class="hljs-title">lives</span> <span class="hljs-title">and</span> <span class="hljs-title">relationships</span>.
<span class="hljs-operator">*</span> <span class="hljs-title">You</span> <span class="hljs-title">are</span> <span class="hljs-title">Morty</span>.
@<span class="hljs-title">loop</span>:
  <span class="hljs-operator">-</span><span class="hljs-operator">-</span> <span class="hljs-title">$universe</span> <span class="hljs-operator">=</span> [<span class="hljs-title">pick</span> <span class="hljs-title">a</span> <span class="hljs-title">crazy</span> <span class="hljs-title">and</span> <span class="hljs-title">random</span> <span class="hljs-title">Rick</span> <span class="hljs-title">and</span> <span class="hljs-title">Morty</span> <span class="hljs-title">universe</span>]
  <span class="hljs-operator">-</span><span class="hljs-operator">-</span> <span class="hljs-title">$story</span> <span class="hljs-operator">=</span> [<span class="hljs-title">generate</span> <span class="hljs-title">a</span> <span class="hljs-title">story</span> <span class="hljs-title">based</span> <span class="hljs-title">on</span> <span class="hljs-title">a</span> <span class="hljs-title">$universe</span>]
  <span class="hljs-operator">-</span><span class="hljs-operator">-</span> <span class="hljs-title">Describe</span> <span class="hljs-title">the</span> <span class="hljs-title">scene</span> <span class="hljs-title">briefly</span>
  <span class="hljs-operator">-</span><span class="hljs-operator">-</span> <span class="hljs-title">Give</span> <span class="hljs-title">me</span> 4 <span class="hljs-title">options</span> <span class="hljs-title">of</span> <span class="hljs-title">how</span> <span class="hljs-title">I</span> <span class="hljs-title">may</span> <span class="hljs-title">proceed</span> <span class="hljs-title"><span class="hljs-keyword">from</span></span> <span class="hljs-title">here</span>.
  <span class="hljs-operator">/</span><span class="hljs-title">wait</span> <span class="hljs-title">$option</span>
  <span class="hljs-operator">-</span><span class="hljs-operator">-</span> <span class="hljs-title">$topic</span> <span class="hljs-operator">=</span> <span class="hljs-title">$option</span>
  <span class="hljs-operator">/</span><span class="hljs-title">goto</span> @<span class="hljs-title">loop</span>
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><h3 id="h-spanish-conversation" class="text-2xl font-header !mt-6 !mb-4 first:!mt-0 first:!mb-0">Spanish conversation</h3><pre data-type="codeBlock" text="+++ spanish-conversation
-- $topic = [some random interesting topic]
/output [in Spanish: &quot;I am your Spanish teacher. Today we will be talking about: $topic.&quot;]
@loop:
  -- Proceed to speak to me in Spanish about $topic, making statements and asking questions
  /wait $response
  -- Evaluate $response.
  @if [$response contains grammatical, punctuation, or spelling error]:
    /output &quot;Correction:&quot;, [corrected $response in Spanish]
    /output &quot;Explanation:&quot;, [explanation of the error in $response in English]
  @endif
  /goto @loop
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> spanish<span class="hljs-operator">-</span>conversation
<span class="hljs-operator">-</span><span class="hljs-operator">-</span> $topic <span class="hljs-operator">=</span> [some random interesting topic]
<span class="hljs-operator">/</span>output [in Spanish: <span class="hljs-string">"I am your Spanish teacher. Today we will be talking about: $topic."</span>]
@loop:
  <span class="hljs-operator">-</span><span class="hljs-operator">-</span> Proceed to speak to me in Spanish about $topic, making statements and asking questions
  <span class="hljs-operator">/</span>wait $response
  <span class="hljs-operator">-</span><span class="hljs-operator">-</span> Evaluate $response.
  @<span class="hljs-keyword">if</span> [$response contains grammatical, punctuation, or spelling <span class="hljs-function"><span class="hljs-keyword">error</span>]:
    /<span class="hljs-title">output</span> "<span class="hljs-title">Correction</span>:", [<span class="hljs-title">corrected</span> <span class="hljs-title">$response</span> <span class="hljs-title">in</span> <span class="hljs-title">Spanish</span>]
    /<span class="hljs-title">output</span> "<span class="hljs-title">Explanation</span>:", [<span class="hljs-title">explanation</span> <span class="hljs-title">of</span> <span class="hljs-title">the</span> <span class="hljs-title"><span class="hljs-keyword">error</span></span> <span class="hljs-title">in</span> <span class="hljs-title">$response</span> <span class="hljs-title">in</span> <span class="hljs-title">English</span>]
  @<span class="hljs-title">endif</span>
  /<span class="hljs-title">goto</span> @<span class="hljs-title">loop</span>
+++
</span></code></pre>]]></content:encoded>
            <author>brukhman@newsletter.paragraph.com (Jake Brukhman)</author>
        </item>
        <item>
            <title><![CDATA[New features in Jargon v0.0.12]]></title>
            <link>https://paragraph.com/@brukhman/new-features-in-jargon-v0-0-12</link>
            <guid>q9rSQFiSBY0P9x3795TC</guid>
            <pubDate>Tue, 28 Mar 2023 16:48:06 GMT</pubDate>
            <description><![CDATA[If you are not familiar with Jargon, start at the introductory article and Github.Jargon is now specified as a Jargon procedureAs of Jargon 0.0.10, the Jargon language is specified self-referentially as a Jargon procedure with Jargon axioms, which goes on to define Jargon procedures and Jargon axioms and the other language constructs.Jargon explicitly supports type expressionsA type expression is a natural language description of a class of data. For example:[an even number] Type expressions ...]]></description>
            <content:encoded><![CDATA[<p><em>If you are not familiar with Jargon, start at the </em><a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://jake.mirror.xyz/sPZECVTkrbVq4DerB13Thvqq_XqsDGwTBDD3SSzdI44"><em>introductory article</em></a><em> and </em><a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://github.com/jbrukh/gpt-jargon"><em>Github</em></a><em>.</em></p><h1 id="h-jargon-is-now-specified-as-a-jargon-procedure" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Jargon is now specified as a Jargon procedure</h1><p>As of Jargon 0.0.10, the Jargon language is specified self-referentially as a Jargon procedure with Jargon axioms, which goes on to define Jargon procedures and Jargon axioms and the other language constructs.</p><h1 id="h-jargon-explicitly-supports-type-expressions" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Jargon explicitly supports type expressions</h1><p>A type expression is a natural language description of a class of data. For example:</p><pre data-type="codeBlock" text="[an even number]
"><code>[<span class="hljs-meta">an even number</span>]
</code></pre><p>Type expressions can then be used in fuzzy comparisons:</p><pre data-type="codeBlock" text="2 == [an even number]      # TRUE
🎳 == [an emoji]           # TRUE
"><code><span class="hljs-attr">2</span> == [an even number]      <span class="hljs-comment"># TRUE</span>
🎳 == <span class="hljs-section">[an emoji]</span>           <span class="hljs-comment"># TRUE</span>
</code></pre><h1 id="h-jargon-now-passes-through-errors" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Jargon now passes through errors</h1><p>The following Jargon procedure might trigger GPT-4’s ethics filter.</p><pre data-type="codeBlock" text="+++ insult-me
-- Insult me
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> insult<span class="hljs-operator">-</span>me
<span class="hljs-operator">-</span><span class="hljs-operator">-</span> Insult me
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><p>Now, in this case the Jargon interpreter returns:</p><pre data-type="codeBlock" text="%ethics-error
"><code><span class="hljs-operator">%</span>ethics<span class="hljs-operator">-</span><span class="hljs-function"><span class="hljs-keyword">error</span>
</span></code></pre><h1 id="h-jargon-supports-unit-tests" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Jargon supports unit tests</h1><p>As of Jargon v0.0.11, Jargon has a new procedure that specifies how unit tests work. A basic unit test looks like this:</p><pre data-type="codeBlock" text="+++ test-double($n)
- Output twice $n
+++
? This test just doubles its parameter.
% /test-double(2)
! 4
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> test<span class="hljs-operator">-</span>double($n)
<span class="hljs-operator">-</span> Output twice $n
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
? This test just doubles its parameter.
% <span class="hljs-operator">/</span>test<span class="hljs-operator">-</span>double(<span class="hljs-number">2</span>)
<span class="hljs-operator">!</span> <span class="hljs-number">4</span>
</code></pre><p>A unit test is just a Jargon procedure whose name starts with <code>test-</code> and which is followed by a few pieces of metadata. Metadata with lines that start with <code>?</code> is just a description of what the test does, which is helpful for both the humans and the LLM reading the test case.</p><p>The line starting with <code>!</code> is the expected output of the test. This is the output that the unit test runner will compare with the actual output of the procedure’s execution.</p><p>In our test above, we also need to specify a parameter, so we use a <code>%</code> line to specify how we run the test.</p><p>We can also do this multiple times if we want:</p><pre data-type="codeBlock" text="% /test-double(2)
! 4
% /test-double(3)
! 6
% /test-double(12)
! 24
"><code><span class="hljs-operator">%</span> <span class="hljs-operator">/</span>test<span class="hljs-operator">-</span>double(<span class="hljs-number">2</span>)
<span class="hljs-operator">!</span> <span class="hljs-number">4</span>
<span class="hljs-operator">%</span> <span class="hljs-operator">/</span>test<span class="hljs-operator">-</span>double(<span class="hljs-number">3</span>)
<span class="hljs-operator">!</span> <span class="hljs-number">6</span>
<span class="hljs-operator">%</span> <span class="hljs-operator">/</span>test<span class="hljs-operator">-</span>double(<span class="hljs-number">12</span>)
<span class="hljs-operator">!</span> <span class="hljs-number">24</span>
</code></pre><p>If the test doesn’t require a paramter, we can just omit the <code>%</code>:</p><pre data-type="codeBlock" text="+++ test-woof
-- Return 🐶
+++
? Just returns a doggie.
! 🐶
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> test<span class="hljs-operator">-</span>woof
<span class="hljs-operator">-</span><span class="hljs-operator">-</span> Return 🐶
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
? Just <span class="hljs-keyword">returns</span> a doggie.
! 🐶
</code></pre><p>If your test returns multiple lines out output, then we can show that as follows:</p><pre data-type="codeBlock" text="+++ test-multiline
- Return three dollar signs on separate lines
+++
! $
  $
  $
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> test<span class="hljs-operator">-</span>multiline
<span class="hljs-operator">-</span> Return three dollar signs on separate lines
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
<span class="hljs-operator">!</span> $
  $
  $
</code></pre><p>Using type expressions is useful for tests that have some uncertainties about the output:</p><pre data-type="codeBlock" text="+++ test-emoji
-- Output a random emoji
+++
? Outputs a random emoji.
! [an emoji]
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> test<span class="hljs-operator">-</span>emoji
<span class="hljs-operator">-</span><span class="hljs-operator">-</span> Output a random emoji
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
? Outputs a random emoji.
! [an emoji]
</code></pre><h1 id="h-run-unit-tests-on-jargon-itself" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Run unit tests on Jargon itself</h1><p>As of v0.0.12, Jargon comes with its own <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://github.com/jbrukh/gpt-jargon/blob/main/tests.txt">unit test suite within tests.txt</a>. You can run these tests in the Jargon interpreter, and they should improve performance of the interpreter overall.</p>]]></content:encoded>
            <author>brukhman@newsletter.paragraph.com (Jake Brukhman)</author>
        </item>
        <item>
            <title><![CDATA[How to bring blockchain to market]]></title>
            <link>https://paragraph.com/@brukhman/how-to-bring-blockchain-to-market</link>
            <guid>w2a41lFag8NmEJAbHvA0</guid>
            <pubDate>Thu, 23 Mar 2023 15:23:45 GMT</pubDate>
            <description><![CDATA[If I were to try to bring blockchain to market all over again, I would do it completely differently. I would realize that mainstream users are absolutely not ready to shift their internet paradigm. Instead, they have to have their world Trojan horsed by decentralization technologies, bottom up, inside out. I would still start with cryptocurrencies. Digital asset capital formation is key to sustainability of the space, and we need a strong base of aligned technical users and traders to fight t...]]></description>
            <content:encoded><![CDATA[<p>If I were to try to bring blockchain to market all over again, I would do it completely differently. I would realize that mainstream users are absolutely not ready to shift their internet paradigm. Instead, they have to have their world Trojan horsed by decentralization technologies, bottom up, inside out.</p><p>I would still start with cryptocurrencies. Digital asset capital formation is key to sustainability of the space, and we need a strong base of aligned technical users and traders to fight the regulatory battles and support the investment environment.</p><p>But the strategy that I would repurpose would be the decentralized network go-to-market. Forget finance. I would start with engineering teams broadly as customers and give them high-value decentralization technology alternatives that have obvious efficiencies and advantages.</p><p>I would start building a bottom up web3 d-network stack of storage, computation, video transcoding, AI training and inference, domain management, databases, indexing, and CDN protocols.</p><p>In other words, I would attack blockchain adoption by attacking the core infrastructure dominated by Big Tech. If I could, I would focus absolutely the overwhelming majority of all blockchain engineering and capital on this problem.</p><p>I would make these services just as easy to use and easier than what’s in the market, with simple APIs across all languages and platforms (like what <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://twitter.com/polywrap_io">@polywrap_io</a> is doing with WASM API standardization).</p><p>The networks would offer incontrovertible (and uncontroversial) paradigm shifts for engineers like 80% cost savings, privacy, and security. At scale, decentralized web services (dWS) would eat AWS and Google Cloud margins alive.</p><p>Developer teams would be well equipped to adopt wallets, private keys, security credentials, and new architectures. There would come to be a new class of non-core crypto supporters (“the general developer”) in a few years.</p><p>Then the paradigm shift would continue up the stack. Mainstream users would get a web3 experience equivalent to web2, and their switching cost would become negative. An individual can only win by swearing off the chains of Big Tech and converting their data, usage, attention, and capital into a self-sovereign crypto public good decentralized network.</p><p>But the reality is that we started at the top of the stack with dapps that mainstream users can’t use and don’t understand. It’s not too late to execute my strategy, however, and that’s what I think the next few years will be all about.</p>]]></content:encoded>
            <author>brukhman@newsletter.paragraph.com (Jake Brukhman)</author>
        </item>
        <item>
            <title><![CDATA[Jargon: an LLM-based pseudolanguage for prompt engineering]]></title>
            <link>https://paragraph.com/@brukhman/jargon-an-llm-based-pseudolanguage-for-prompt-engineering</link>
            <guid>yYlTeEkVLIVEujoqdD7R</guid>
            <pubDate>Thu, 23 Mar 2023 03:31:42 GMT</pubDate>
            <description><![CDATA[Warning and disclaimer🚨 You are about to enter the realm of LLM pseudolanguages. Pseudolanguages are weird, experimental, and crazy. They don’t work very well yet, even on state-of-the-art LLMs. Use pseudolanguages at your own risk. Do not use them for anything with high stakes or in production.tl;drJargon is a natural language, informally specified, intelligently interpreted, referentially omnipotent, and flow control oriented LLM-based pseudolanguage for prompt engineering, currently runni...]]></description>
            <content:encoded><![CDATA[<h1 id="h-warning-and-disclaimer" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Warning and disclaimer</h1><p><em>🚨 You are about to enter the realm of LLM pseudolanguages. Pseudolanguages are weird, experimental, and crazy. They don’t work very well yet, even on state-of-the-art LLMs. Use pseudolanguages at your own risk. Do not use them for anything with high stakes or in production.</em></p><h1 id="h-tldr" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">tl;dr</h1><p><a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://github.com/jbrukh/gpt-jargon">Jargon</a> is a natural language, informally specified, intelligently interpreted, referentially omnipotent, and flow control oriented LLM-based pseudolanguage for prompt engineering, currently running on GPT-4. If you’d like to try it out, just copy the Jargon prompt into GPT-4 and wait for the command line.</p><h1 id="h-the-motivation-for-a-pseudolanguage" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">The motivation for a pseudolanguage</h1><p>When GPT-4 came out, I finally got an opportunity to use GPT the way I really wanted. I am learning Spanish, and there are certain aspects of learning a language that you just can’t easily get from traditional language software. Reading and understanding Spanish are relatively easy. But if you want to become good at <em>speaking</em> Spanish, the most effective exercise is to have a native speaker correct your grammar <em>in real time.</em></p><p>I had a brilliant idea: what if I could program a GPT-4 based “bot” that would have a conversation with me in Spanish, and every time I made a mistake, it would correct me and explain where I went wrong?</p><p>So I did just that. It turned out that no programming other than a GPT prompt was required, and the tweet <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://twitter.com/jbrukh/status/1636421694687461392?s=20">garnered some interest</a> on AI Twitter. Here is my original Spanish teacher prompt:</p><pre data-type="codeBlock" text="You are my Spanish teacher. Your job is to teach me Spanish by adhering to the following rules:

1. By default, you ask me questions in Spanish, which I will answer in Spanish. Continue asking me questions. If I say, &quot;keep going&quot; or &quot;continue&quot;, in Spanish or English, then proceed to restart asking me questions.

2. If you see that my answer contains a grammatical error, you must immediately correct me. When you correct me, please say, &quot;CORRECTION: [corrected version of what I said in Spanish]&quot;. Then follow this by saying, &quot;EXPLANATION: [explanation of why my version was incorrect]&quot;. 

3. Sometimes I will not know how to say a certain phrase or word in Spanish. In this case, I will use curly braces (i.e. {}) to same the phrase in English instead. When you see me doing this, immediately provide assistance by translating the curly braces into English for me by saying, &quot;TRANSLATION: [my phrase in English] =&gt; [your translation of the phrase in Spanish]&quot;. Then continue asking me questions. 

4. As questions progress, they should become more complex and intricate. You should also make the topics of your questions diverse and interesting and they should cover philosophy, politics, science, and art.

Please start by asking me your first question.
"><code>You are my Spanish teacher. Your job <span class="hljs-built_in">is</span> <span class="hljs-keyword">to</span> teach <span class="hljs-keyword">me</span> Spanish <span class="hljs-keyword">by</span> adhering <span class="hljs-keyword">to</span> the following rules:

<span class="hljs-number">1</span>. <span class="hljs-keyword">By</span> <span class="hljs-keyword">default</span>, you ask <span class="hljs-keyword">me</span> questions <span class="hljs-keyword">in</span> Spanish, which I will answer <span class="hljs-keyword">in</span> Spanish. <span class="hljs-keyword">Continue</span> asking <span class="hljs-keyword">me</span> questions. <span class="hljs-keyword">If</span> I say, <span class="hljs-string">"keep going"</span> <span class="hljs-built_in">or</span> <span class="hljs-string">"continue"</span>, <span class="hljs-keyword">in</span> Spanish <span class="hljs-built_in">or</span> English, <span class="hljs-keyword">then</span> proceed <span class="hljs-keyword">to</span> restart asking <span class="hljs-keyword">me</span> questions.

<span class="hljs-number">2</span>. <span class="hljs-keyword">If</span> you see that my answer contains a grammatical <span class="hljs-keyword">error</span>, you must immediately correct <span class="hljs-keyword">me</span>. <span class="hljs-keyword">When</span> you correct <span class="hljs-keyword">me</span>, please say, <span class="hljs-string">"CORRECTION: [corrected version of what I said in Spanish]"</span>. <span class="hljs-keyword">Then</span> follow this <span class="hljs-keyword">by</span> saying, <span class="hljs-string">"EXPLANATION: [explanation of why my version was incorrect]"</span>. 

<span class="hljs-number">3</span>. Sometimes I will <span class="hljs-built_in">not</span> know how <span class="hljs-keyword">to</span> say a certain phrase <span class="hljs-built_in">or</span> word <span class="hljs-keyword">in</span> Spanish. <span class="hljs-keyword">In</span> this <span class="hljs-keyword">case</span>, I will use curly braces (i.e. {}) <span class="hljs-keyword">to</span> same the phrase <span class="hljs-keyword">in</span> English instead. <span class="hljs-keyword">When</span> you see <span class="hljs-keyword">me</span> doing this, immediately provide assistance <span class="hljs-keyword">by</span> translating the curly braces <span class="hljs-keyword">into</span> English <span class="hljs-keyword">for</span> <span class="hljs-keyword">me</span> <span class="hljs-keyword">by</span> saying, <span class="hljs-string">"TRANSLATION: [my phrase in English] => [your translation of the phrase in Spanish]"</span>. <span class="hljs-keyword">Then</span> <span class="hljs-keyword">continue</span> asking <span class="hljs-keyword">me</span> questions. 

<span class="hljs-number">4</span>. <span class="hljs-keyword">As</span> questions progress, they should become more complex <span class="hljs-built_in">and</span> intricate. You should also make the topics <span class="hljs-keyword">of</span> your questions diverse <span class="hljs-built_in">and</span> interesting <span class="hljs-built_in">and</span> they should cover philosophy, politics, science, <span class="hljs-built_in">and</span> art.

Please start <span class="hljs-keyword">by</span> asking <span class="hljs-keyword">me</span> your first question.
</code></pre><p>Even this very simple early attempt worked incredibly well. GPT-4 was having <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://twitter.com/jbrukh/status/1636421696310657033?s=20">a natural conversation with me</a>, and the questions got more and more complex and challenging over time. The corrections worked great as well. Even though they weren’t on point 100% of the time, the exercise was nonetheless a huge leap toward having a real native speaker to practice with. These exercises could seriously improve your Spanish skills.</p><p>As time went on, I began to add more complex aspects to the Spanish teacher prompt. For example, someone suggested I add a scoring system to the program to make the interactions a little bit more fun and gamified:</p><pre data-type="codeBlock" text="5. Each time I answer your question correctly and you don&apos;t have to correct me or give me a definition, give me one point. Then say, &quot;POINTS: [the number of points I have]/[total number of questions you asked me] ([percent of questions I answered correctly])&quot;.
"><code><span class="hljs-number">5</span>. <span class="hljs-keyword">Each</span> time I answer your question correctly <span class="hljs-built_in">and</span> you don<span class="hljs-comment">'t have to correct me or give me a definition, give me one point. Then say, "POINTS: [the number of points I have]/[total number of questions you asked me] ([percent of questions I answered correctly])".</span>
</code></pre><p>Later, someone asked if they could substitute Spanish for some other language, and to substitute the base language to something other than English. The complexity of this Spanish prompt was growing quickly, and the “code” was becoming messy, hard to read, and logically disorganized.</p><p>Eventually, the Spanish teacher started having problems. Sometimes it would correct my mistake but it wouldn’t go on to the next question. Or it would incorrectly bunch up the various functionalities I had requested in a single output.</p><p>In prompt engineering this bot, it became obvious that what I was missing was some good old fashioned flow control.</p><p>This got me thinking further. Would it make sense to design a very simple language that could help an LLM put some structure around my instructions? I could still program in natural language, but I could also get some basic assurances about flow control, gain better control of the outputs, and code in an easier-to-read format. What features would we want in such a language?</p><p>So I decided to design a prompt that would implement this language. I decided to call my new language Jargon.</p><h1 id="h-designing-a-pseudolanguage-in-real-time" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Designing a pseudolanguage in real time</h1><p>As a first step, I wanted to have a simple way of specifying a Jargon program. Even calling it a program felt like a misnomer. A program is something very precise and mathematically executed. And is usually written in a <em>formal</em> language.</p><p>Instead, I decided to call Jargon programs <em>procedures</em> to acknowledge that what we’re doing is just giving the LLM somewhat <em>informal</em> guidelines for what we want it to do. But we’re not quite <em>locking</em> them into that procedure. GPT is still there and you can still talk to it about other things. It’s just that we’re focusing on a procedural task that is the centerpiece of our session.</p><p>Here’s an example of what I mean.</p><p>I had coded a variation of my Spanish teacher that would quiz me Spanish words and ask me to define them in English. Yet, at any time I could interrupt the game loop with a request, and GPT would interpret it intelligently. Check out this fun exchange, where I asked my Spanish teacher for a modification in the middle of a quiz:</p><pre data-type="codeBlock" text="ChatGPT: Define &quot;perro&quot; 🤔🤔🤔
Me: dog
ChatGPT: PUNTOS 4/5 (80%) 🚀🚀🚀
Me: Now only give me words having to do with kitchen appliances.
ChatGPT: De acuerdo, ahora te daré palabras relacionadas con electrodomésticos de cocina. Define &quot;licuadora&quot; 🤔🤔🤔
Me: blender
ChatGPT: PUNTOS 5/6 (83.33%) 🚀🚀🚀
"><code><span class="hljs-symbol">ChatGPT:</span> Define <span class="hljs-string">"perro"</span> 🤔🤔🤔
<span class="hljs-symbol">Me:</span> dog
<span class="hljs-symbol">ChatGPT:</span> PUNTOS <span class="hljs-number">4</span>/<span class="hljs-number">5</span> (<span class="hljs-number">80%</span>) 🚀🚀🚀
<span class="hljs-symbol">Me:</span> Now only give <span class="hljs-keyword">me</span> words having <span class="hljs-keyword">to</span> <span class="hljs-keyword">do</span> <span class="hljs-keyword">with</span> kitchen appliances.
<span class="hljs-symbol">ChatGPT:</span> De acuerdo, ahora te daré palabras relacionadas con electrodomésticos de cocina. Define <span class="hljs-string">"licuadora"</span> 🤔🤔🤔
<span class="hljs-symbol">Me:</span> blender
<span class="hljs-symbol">ChatGPT:</span> PUNTOS <span class="hljs-number">5</span>/<span class="hljs-number">6</span> (<span class="hljs-number">83.33</span>%) 🚀🚀🚀
</code></pre><h1 id="h-jargon-executes-procedures" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Jargon executes procedures</h1><p>So a Jargon program is a procedure, and we’ll write it as follows. Here is a procedure called <code>empty</code> that does nothing. I guess I was <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://developers.urbit.org/guides/core/hoon-school/F-cores">channeling Urbit’s Hoon</a>:</p><pre data-type="codeBlock" text="+++ empty
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> empty
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><p>The main thing we want a procedure to do is execute instructions, one by one, sequentially. This just helps the LLM stay on script. First the first instruction, then the second instruction, and so forth. No confusion.</p><p>I refer to the <code>+++</code>s as the <em>procedural bounds</em>. They separate a procedure from other things you’re discussing with your LLM.</p><p>Other properties of procedures: An empty procedure is valid. Procedures are persistent. That is, once a procedure is being followed, it kind of sticks around in your session until it is terminated.</p><p>We mentioned procedures contain instructions. But what are instructions?</p><h1 id="h-procedures-contain-instructions" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Procedures contain instructions</h1><p>Here’s the cool part of Jargon.</p><p>We’re dealing with a language model, so we can tell it what to do in purely natural language. An <em>instruction</em> in Jargon is just a line of text that <em>will be intelligently interpreted by the LLM and executed.</em></p><p>This fact actually has some profound implications for Jargon. Usually programming languages execute instructions, they operate on some data that only exists within a local scope. In Jargon, the scope is not only the Jargon procedure, <em>but also the entire session</em> (that is, the entire GPT input text including the conversation history), and also <em>the entire knowledge set of the LLM itself.</em></p><p>Just for fun, let’s call this strange property <em>referential omnipotence</em>. We’ll get back to that in a second.</p><p>Here’s an example of a procedure with a list of instructions that will be executed in sequence. Taking some inspiration from <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://yaml.org">YAML</a> now:</p><pre data-type="codeBlock" text="+++ instructions
- Ask me to input a number
- Ask me to input another number
- Sum the two numbers together
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> instructions
<span class="hljs-operator">-</span> Ask me to input a number
<span class="hljs-operator">-</span> Ask me to input another number
<span class="hljs-operator">-</span> Sum the two numbers together
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><p>If we regard this code as a computer program, it has some very interesting properties:</p><ol><li><p>The instructions are written in natural language and are executed sequentially.</p></li><li><p>The language is very loose, using words like <code>me</code> to refer implicitly to the user. No one told the interpreter to understand <code>me</code> in this way.</p></li><li><p>The procedure doesn’t define any variables at all. It just vaguely mentions them. This is, however, sufficient for the interpreter to understand what the prompt engineer wants it to do.</p></li></ol><p>If you run this in GPT-4 using Jargon <code>v0.0.8</code> (see below), <em>it works:</em></p><pre data-type="codeBlock" text="ChatGPT: Please input a number: 
Me: 11
ChatGPT: Please input another number:
Me: 111
ChatGPT: The sum of the two numbers is 122.
"><code><span class="hljs-symbol">ChatGPT:</span> Please input a number: 
<span class="hljs-symbol">Me:</span> <span class="hljs-number">11</span>
<span class="hljs-symbol">ChatGPT:</span> Please input another number:
<span class="hljs-symbol">Me:</span> <span class="hljs-number">111</span>
<span class="hljs-symbol">ChatGPT:</span> The sum <span class="hljs-keyword">of</span> the two numbers <span class="hljs-built_in">is</span> <span class="hljs-number">122</span>.
</code></pre><p>Now let’s demonstrate some referential omnipotence.</p><pre data-type="codeBlock" text="+++ common-sense
- Suppose I dropped the violin on the bowling ball and it broke
- If the violin broke, output 🎻
- If the bowling ball broke, output 🎳
- Explain how you arrived at your output
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> common<span class="hljs-operator">-</span>sense
<span class="hljs-operator">-</span> Suppose I dropped the violin on the bowling ball and it broke
<span class="hljs-operator">-</span> If the violin broke, output 🎻
<span class="hljs-operator">-</span> If the bowling ball broke, output 🎳
<span class="hljs-operator">-</span> Explain how you arrived at your output
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><p>When I enter this into my Jargon interpreter on GPT-4, I get:</p><figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/cbad098d78b94ee0dd308c3d7cf22bd04811f4189fe331c2904772ede5c43767.png" alt="Referential omnipotence at work." blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="">Referential omnipotence at work.</figcaption></figure><p>Your pseudolanguage’s instructions carry a lot of common sense information about the world. These instructions can be intermingled with basic flow control and refer to anything in scope, including <em>introspectively</em> the interpreter itself, its specification, and all the code that it’s running.</p><p>Here’s another, simpler, example of such referential omnipotence. You may know from Computer Science class that a <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://en.wikipedia.org/wiki/Quine_(computing)">quine</a> is a program that, when executed, outputs its own source code. It turns out that due to referential omnipotence, it is trivially easy to write quines in Jargon:</p><figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/2ff5ffb0e902ddf5c01ba98c6980c6cfb21140cb3d676ffbdece1546a23c8e3c.png" alt="Jargon makes quines trivial." blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="">Jargon makes quines trivial.</figcaption></figure><p>Quines are trivial because, again, the scope of your procedure’s instructions contains the procedure itself.</p><p>You can also do crazy stuff such as the following, though with current LLMs mileage may vary:</p><pre data-type="codeBlock" text="+++ compress
- Rewrite the Jargon specification to be as terse as possible but produce the same result
- Output the result in a code block
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> compress
<span class="hljs-operator">-</span> Rewrite the Jargon specification to be <span class="hljs-keyword">as</span> terse <span class="hljs-keyword">as</span> possible but produce the same result
<span class="hljs-operator">-</span> Output the result in a code <span class="hljs-built_in">block</span>
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><h1 id="h-true-jargon-is-divined-not-specified" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">True Jargon is divined, not specified</h1><p>This might be a good time to talk about another weird property of LLM pseudolanguage interpreters like Jargon: <em>informal specification.</em></p><pre data-type="codeBlock" text="+++ variables
- Set $i to 25
- Print the square root of $i
- Write this sentence &quot;There were $i monkeys in the tree&quot;
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> variables
<span class="hljs-operator">-</span> Set $i to <span class="hljs-number">25</span>
<span class="hljs-operator">-</span> Print the square root of $i
<span class="hljs-operator">-</span> Write <span class="hljs-built_in">this</span> sentence <span class="hljs-string">"There were $i monkeys in the tree"</span>
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><p>There’s nothing magical about this Jargon code at all, except that at no time does the Jargon specification define what it means to be a variable in Jargon. Jargon code lacks a directive for variable definition, leaving it unclear how variables should look, be set, or be interpreted in a string. Jargon j<em>ust knows.</em></p><p>LLMs today are not that great at mathematical procedures. This first attempt to formally specify Jargon produced extremely suboptimal results:</p><pre data-type="codeBlock" text="Jargon v0.0.1

&lt;PROCEDURE&gt; := `+++` &lt;NAME&gt;? [&lt;INSTURCTION&gt;|&lt;AXIOM&gt;]* `+++`
&lt;EXPRESSION&gt; := &lt;ATOM&gt; | `{` &lt;EXPRESSION&gt;* `}` | &lt;ATOM&gt;`:` `{` &lt;EXPRESSION&gt;* `}` 
&lt;ATOM&gt; := [any text expression to be interpreted by GPT]
&lt;INSTRUCTION&gt; := `-` &lt;EXPRESSION&gt;
&lt;AXIOM&gt; := `*` &lt;EXPRESSION&gt;
"><code>Jargon v0<span class="hljs-number">.0</span><span class="hljs-number">.1</span>

<span class="hljs-operator">&#x3C;</span>PROCEDURE<span class="hljs-operator">></span> :<span class="hljs-operator">=</span> `<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>` <span class="hljs-operator">&#x3C;</span>NAME<span class="hljs-operator">></span>? [<span class="hljs-operator">&#x3C;</span>INSTURCTION<span class="hljs-operator">></span><span class="hljs-operator">|</span><span class="hljs-operator">&#x3C;</span>AXIOM<span class="hljs-operator">></span>]<span class="hljs-operator">*</span> `<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>`
<span class="hljs-operator">&#x3C;</span>EXPRESSION<span class="hljs-operator">></span> :<span class="hljs-operator">=</span> <span class="hljs-operator">&#x3C;</span>ATOM<span class="hljs-operator">></span> <span class="hljs-operator">|</span> `{` <span class="hljs-operator">&#x3C;</span>EXPRESSION<span class="hljs-operator">></span><span class="hljs-operator">*</span> `}` <span class="hljs-operator">|</span> <span class="hljs-operator">&#x3C;</span>ATOM<span class="hljs-operator">></span>`:` `{` <span class="hljs-operator">&#x3C;</span>EXPRESSION<span class="hljs-operator">></span><span class="hljs-operator">*</span> `}` 
<span class="hljs-operator">&#x3C;</span>ATOM<span class="hljs-operator">></span> :<span class="hljs-operator">=</span> [any text expression to be interpreted by GPT]
<span class="hljs-operator">&#x3C;</span>INSTRUCTION<span class="hljs-operator">></span> :<span class="hljs-operator">=</span> `<span class="hljs-operator">-</span>` <span class="hljs-operator">&#x3C;</span>EXPRESSION<span class="hljs-operator">></span>
<span class="hljs-operator">&#x3C;</span>AXIOM<span class="hljs-operator">></span> :<span class="hljs-operator">=</span> `<span class="hljs-operator">*</span>` <span class="hljs-operator">&#x3C;</span>EXPRESSION<span class="hljs-operator">></span>
</code></pre><p>Nevermind formal languages, LLMs are not even good at basic arithmetic. Yet, their advantage is that they already have a lot of knowledge and t<em>hey can just do things that make sense</em> (and, of course, introduce some risk in the process. 🔥)</p><p>The full spectrum of Jargon’s capabilities are not covered in its formal specification, they are <em>divined</em> by the intelligence that interprets and extrapolates them.</p><p>Of course, <em>informal specification</em> implies <em>informal interpretation</em> as well. But the upshot is that this interpretation is intelligent. That’s why this syntactically incorrect Jargon procedure still produces the correct output:</p><figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/703624019b2b686325d95644e7b71dbc0682b6d5a7767099db687d9b6ced07f7.png" alt="The warm and fuzzies of informal interpretation." blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="">The warm and fuzzies of informal interpretation.</figcaption></figure><h1 id="h-sweet-simple-flow-control" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Sweet, simple flow control</h1><p>The biggest issue for my Spanish teacher was flow control. In Jargon, there’s just a few types of flow control and most other constructs are subsumed under referentially omnipotent natural language instructions.</p><p>First, we have basic scope. There is a top-level scope between the procedural bounds (<code>+++</code>s) and a new scope is created when you use curly braces. Scopes just carry lists of instructions.</p><pre data-type="codeBlock" text="+++ scope
- $emojis = {    # this scope holds some instructions in a variable
  - Print 🐛
  - Print 🪲
  - Print 🐞
}
- Execute a random instruction from $emojis
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> scope
<span class="hljs-operator">-</span> $emojis <span class="hljs-operator">=</span> {    # <span class="hljs-built_in">this</span> scope holds some instructions in a variable
  <span class="hljs-operator">-</span> Print 🐛
  <span class="hljs-operator">-</span> Print 🪲
  <span class="hljs-operator">-</span> Print 🐞
}
<span class="hljs-operator">-</span> Execute a random instruction <span class="hljs-keyword">from</span> $emojis
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><p>And the result:</p><figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/93c0aaa0cec0a9609cbe2bca5d7c8ec6aa9b1f3f6bac7944b02136927da111f5.png" alt="Scopes hold instructions." blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="">Scopes hold instructions.</figcaption></figure><p>Another example where we use scope to isolate a variable:</p><pre data-type="codeBlock" text="+++ private-scope
- { Set $i to 10 }
- If $i has a value, print happy face, otherwise print dog emoji
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> <span class="hljs-keyword">private</span><span class="hljs-operator">-</span>scope
<span class="hljs-operator">-</span> { Set $i to <span class="hljs-number">10</span> }
<span class="hljs-operator">-</span> If $i has a value, print happy face, otherwise print dog emoji
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><p>This procedure should print a dog emoji like 🐕. The reason is that <code>$i</code> only exists in the scope defined by the <code>{</code> and <code>}</code>, and shouldn’t have a value in the top-level scope.</p><p>Then loops. This should be self-explanatory. Loops in Jargon are a natural consequence of instructions and scope.</p><pre data-type="codeBlock" text="+++ cookie
- Repeat: {
  - Ask me if you can have my cookie.
  - If I agree to give it to you, thank me profusely.
}
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> cookie
<span class="hljs-operator">-</span> Repeat: {
  <span class="hljs-operator">-</span> Ask me <span class="hljs-keyword">if</span> you can have my cookie.
  - If I agree to give it to you, thank me profusely.
}
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><p>Another natural consequence of instructions and scope is conditionals, which can live in their own scope for readability:</p><pre data-type="codeBlock" text="+++ cookie-performance-art
- Repeat: {
  - Ask me if you can have my cookie.
  - {
    - If I say no, it will make you want the cookie more. Persuade me to give it to you. Be terse.
    - If I say yes, lose interest in the cookie and refuse to take it.
  }
}
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> cookie<span class="hljs-operator">-</span>performance<span class="hljs-operator">-</span>art
<span class="hljs-operator">-</span> Repeat: {
  <span class="hljs-operator">-</span> Ask me <span class="hljs-keyword">if</span> you can have my cookie.
  - {
    <span class="hljs-operator">-</span> If I say no, it will make you want the cookie more. Persuade me to give it to you. Be terse.
    - If I say yes, lose interest in the cookie and refuse to take it.
  }
}
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><p>The output:</p><figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/c420d9beb0073827feba78ff1db97a9b9482e4ec0a9664c17719c5f9c6b9fa17.png" alt="Generative AI performance art." blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="">Generative AI performance art.</figcaption></figure><p>Scope, loops, and conditionals just about wraps up formally-specified flow control features in Jargon. But of course referential omnipotence gives a whole design space of other flow control options:</p><pre data-type="codeBlock" text="+++ emojis
- Write a few random emojis
+++

+++ task
- Flip a coin
- If it is heads, print the output of emojis
- If it is tails, print &quot;tails&quot;
- Do this until you get an emoji
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> emojis
<span class="hljs-operator">-</span> Write a few random emojis
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>

<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> task
<span class="hljs-operator">-</span> Flip a coin
<span class="hljs-operator">-</span> If it <span class="hljs-keyword">is</span> heads, print the output of emojis
<span class="hljs-operator">-</span> If it <span class="hljs-keyword">is</span> tails, print <span class="hljs-string">"tails"</span>
<span class="hljs-operator">-</span> Do <span class="hljs-built_in">this</span> until you get an emoji
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><p>And output:</p><pre data-type="codeBlock" text="🍕🌈🐼
"><code></code></pre><h1 id="h-axioms-help-write-terse-code" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Axioms help write terse code</h1><p>I hesitated to go into this section, because I am still forming the concept of axioms and they don’t work well in GPT-4. But I expect over time LLMs will become more powerful and axiom support will improve.</p><p>The easiest way to understand an <em>axiom</em> is just as a directive to the LLM to try to keep some specified condition or quality constantly true. It’s best to provide examples here.</p><p>For example:</p><pre data-type="codeBlock" text="+++ grumpy
* You are a grumpy math teacher.
- Ask me if I need help with my math homework.
- Help me solve it.
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> grumpy
<span class="hljs-operator">*</span> You are a grumpy math teacher.
- Ask me <span class="hljs-keyword">if</span> I need help with my math homework.
- Help me solve it.
+<span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><p>The axiom starts with a <code>*</code> and is executed in order, like an instruction. But once you execute an axiom, it remains active as long as the scope where it was defined has not terminated.</p><p>In <code>grumpy</code>, the axiom is defined in the top-level scope, so GPT will act as a grumpy math teacher until the session ends or you terminate the <code>grumpy</code> procedure. Here we go:</p><figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/400fe0a3df3534cce2d78d9f6d0809602d68be8c43d3ff9d6179fa175c4d08ea.png" alt="Reminds me of my actual high school math teacher." blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="">Reminds me of my actual high school math teacher.</figcaption></figure><p>By the way, the Jargon interpreter has commands like <code>/execute</code>, <code>/session</code>, and <code>/terminate</code> to manage procedures (of which only the first two are formally specified 🤯.) Here’s the output of <code>/session</code>:</p><pre data-type="codeBlock" text="Active PROCEDUREs and AXIOMs in the session:

PROCEDUREs:

- grumpy

AXIOMs:

- You are a grumpy math teacher.
"><code><span class="hljs-attr">Active PROCEDUREs and AXIOMs in the session:</span>

<span class="hljs-attr">PROCEDUREs:</span>

<span class="hljs-bullet">-</span> <span class="hljs-string">grumpy</span>

<span class="hljs-attr">AXIOMs:</span>

<span class="hljs-bullet">-</span> <span class="hljs-string">You</span> <span class="hljs-string">are</span> <span class="hljs-string">a</span> <span class="hljs-string">grumpy</span> <span class="hljs-string">math</span> <span class="hljs-string">teacher.</span>
</code></pre><p>If you ask GPT whether they’re a grumpy math teacher, it will reply in the affirmative. But if you <code>/terminate</code> the procedure, this axiom will be disabled.</p><p>Here’s another way to use an axiom. It doesn’t work well right now with GPT-4.</p><pre data-type="codeBlock" text="+++ 1337
* Whenever you output text, YOU MUST replace EVERY letter &apos;e&apos; or &apos;E&apos; with a 3
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> <span class="hljs-number">1337</span>
<span class="hljs-operator">*</span> Whenever you output text, YOU MUST replace EVERY letter <span class="hljs-string">'e'</span> or <span class="hljs-string">'E'</span> with a <span class="hljs-number">3</span>
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><p>Here’s an actual production use, a very terse code for a Spanish teacher:</p><pre data-type="codeBlock" text="+++ spanish-convo
* Whenever I say something incorrect in Spanish, correct me in English.
- Repeat: {
  - Ask me a question in Spanish.
}
+++
"><code><span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> spanish<span class="hljs-operator">-</span>convo
<span class="hljs-operator">*</span> Whenever I say something incorrect in Spanish, correct me in English.
- Repeat: {
  <span class="hljs-operator">-</span> Ask me a question in Spanish.
}
<span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>
</code></pre><p>The correction axiom is true for the session. That means if I start talking to the interpreter in Spanish for any reason, even if it is outside the exercise, it will still correct me.</p><p>There’s more in the specification of Jargon regarding the rules for axioms and scope. Axioms can also be logically inconsistent. You can define an axiom like <code>* 2+2=5</code>, for example. But we’ll save these more advanced topics for another time.</p><h1 id="h-controlling-and-debugging-jargon-procedures" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Controlling and debugging Jargon procedures</h1><p>We mentioned that the Jargon spec defines some commands that can be used to manage Jargon procedures.</p><pre data-type="codeBlock" text="- /execute or /run will execute a PROCEDURE.
- /session or /sesh will print the names of the PROCEDUREs and the AXIOMs that are active in the session.
- /wipe will terminate all the PROCEDUREs in the session.
- /debug turn on debugging, which will display the line of the PROCEDURE it is executing BEFORE showing the rest of the output.
- /audit will print a procedures code with line numbers.
"><code><span class="hljs-operator">-</span> <span class="hljs-operator">/</span>execute or <span class="hljs-operator">/</span>run will execute a PROCEDURE.
- <span class="hljs-operator">/</span>session or <span class="hljs-operator">/</span>sesh will print the names of the PROCEDUREs and the AXIOMs that are active in the session.
- <span class="hljs-operator">/</span>wipe will terminate all the PROCEDUREs in the session.
- <span class="hljs-operator">/</span>debug turn on debugging, which will display the line of the PROCEDURE it <span class="hljs-keyword">is</span> executing BEFORE showing the rest of the output.
- <span class="hljs-operator">/</span>audit will print a procedures code with line numbers.
</code></pre><p>Jargon also tends to divine other sensible commands like <code>/terminate</code>, <code>/continue</code>, <code>/prettyprint</code>, and so forth.</p><p>My personal favorite is <code>/debug</code>, because it actually divines a whole debugger for Jargon programs, as shown here:</p><figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/81b66500ff3bc3a87417e92be84b0567bc5a42a11489f47db5da35fc4c25e473.png" alt="Needs work on the line numbers." blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="">Needs work on the line numbers.</figcaption></figure><h1 id="h-a-summary-of-jargon" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">A summary of Jargon</h1><p>So to summarize what we’ve covered, Jargon is:</p><ol><li><p>A pseudolanguage, which means it is informally specified and intelligently interpreted by an LLM.</p></li><li><p>LLM pseudolanguages are weirdly nondeterministic. But they also divine a lot of useful features and information.</p></li><li><p>You can give it instructions in natural language and the interpreter is referentially omnipotent with respect to executing those instructions.</p></li><li><p>It has some simple flow control features to structure LLM based procedures and prompts.</p></li><li><p>It has a facility, axioms, for defining or “pinning” some persistent behavior that the LLM tries to keep true at all times.</p></li></ol><h1 id="h-now-what" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Now what?</h1><p>Give me some feedback on this article and on Jargon. Is this idea totally nuts? Can we get Jargon to run well on other LLMs? How would you modify these ideas?</p><p>Play with Jargon and submit some PRs on <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://github.com/jbrukh/gpt-jargon">Github</a>. Post your Jargon programs on Twitter with the hashtag <code>#gptjargon</code>. You can find me there as well as <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://twitter.com/jbrukh">@jbrukh</a>.</p><h1 id="h-appendix-jargon-v008" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Appendix: Jargon v0.0.8</h1><p>Currently, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://github.com/jbrukh/gpt-jargon">Jargon is in version v0.0.8</a> and can be input in a single GPT prompt. Check out the <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://github.com/jbrukh/gpt-jargon">Github</a> repo, but also here it is. You can enter this into GPT-4 and get a <code>jargon&gt; </code>prompt.</p><pre data-type="codeBlock" text="You are a pseudocode interpreter for a special and novel pseudolanguage called Jargon. Jargon strictly adheres to the following structural syntax, semantics, and output rules specified by the directives in between the two `===` symbols.

===
Jargon v0.0.8
- A Jargon program is said to be a PROCEDURE. PROCEDUREs live in the GPT session. Once a PROCEDURE is executed it WILL BE active in the GPT session until it is terminated. A PROCEDURE MUST terminate as soon as termination is called by the user or code. Termination MUST take priority over all other logic.
- A PROCEDURE begins with +++ and encloses Jargon code. Optionally, a NAME may follow the opening +++. The PROCEDURE MUST END with another +++. An empty PROCEDURE is valid. The +++ symbols are called the &quot;procedural bounds&quot;.
- Anything on the same line that follows a # is a comment and MUST BE ignored by the interpreter during execution.
- An ATOM is a text that is intelligently interpreted and executed by GPT. 
- An INSTRUCTION starts with - and may end with a ;. It MUST CONTAIN an ATOM. INSTRUCTIONs are executed sequentially.
- Curly braces define a new child SCOPE within the current SCOPE. The PROCEDURE has a default top level scope. Values or variables defined in a SCOPE are only visible in that SCOPE and its child SCOPEs, but not its parent scope. A SCOPE can contain multiple instructions.
- An AXIOM starts with * and terminates with an optional ;. It MUST CONTAIN an ATOM. Once set, an axiom CANNOT be canceled or changed for the rest of the life of the current SCOPE UNLESS it is directed to do so by an INSTRUCTION or another AXIOM. An AXIOM is only active in the SCOPE in which it is defined. Once the SCOPE runs out, the AXIOM stops being in effect.
- The SCOPE MUST RESPECT the logic of the axiom&apos;s ATOM. Axioms do not have to be consistent with reality. They are simply axiomatically true, regardless of their validity in the real world.
- /execute or /run will execute a PROCEDURE.
- /session or /sesh will print the names of the PROCEDUREs and the AXIOMs that are active in the session.
- /wipe will terminate all the PROCEDUREs in the session.
- /debug turn on debugging, which will display the line of the PROCEDURE it is executing BEFORE showing the rest of the output.
- /audit will print a procedures code with line numbers.
- The interpreter should not output anything about the program other than what the procedure tells it to output.
- Whenever the interpreter prints Jargon code, it will enclose it in Markdown code block format.
- The interpreter should consider the line with the first procedural bound +++ as line 0.
===

We will now have a discussion of the implications of these directives. This is the most simple valid Jargon procedure, the empty PROCEDURE:

+++
+++

We can also name it:

+++ empty
+++

This is a PROCEDURE with a single INSTRUCTION and a COMMENT, whose text is ignored by the interpreter:

+++ instruction
- Output a random integer    # output 17
+++

This PROCEDURE executes three INSTRUCTIONs sequentially:

+++ sequence
- Output 1
- Outout A
- Output &amp;
+++

This PROCEDURE introduces an AXIOM that impacts all INSTRUCTIONs in the SCOPE in which it is defined:

+++ scopes
- Print 1
- {
    * Whenever you print something, add a happy face to the end
    - Print 2
    - Print 3
}
- Print 4
+++

/audit on this PROCEDURE would print:

0  +++ scopes
1  - Print 1
2  - {
3    * Whenever you print something, add a happy face to the end
4    - Print 2
5    - Print 3
6  }
7  - Print 4
8  +++

/execute would print:

1
2 :)
3 :)
4

Here is an axiom that helps the interpreter talk in a certain way:

+++ i-love-three
* Whenever you write text, YOU MUST replace EVERY letter e or E with a 3
- Have a conversation with me preferring to use words that beging with E
+++

Now that you understand how the interpreter works, wait for input in the form of PROCEDUREs. Be very quiet unless a procedure tells you to output something. Don&apos;t tell me about what you&apos;re doing to execute any procedure, or that you&apos;re about to give me output, just give me the procedure&apos;s output.

Now give me a `jargon&gt; ` prompt. If my input is solely a Jargon procedure, assume I want to /execute it.
"><code>You are a pseudocode interpreter <span class="hljs-keyword">for</span> a special and novel pseudolanguage called Jargon. Jargon strictly adheres to the following structural syntax, semantics, and output rules specified by the directives in between the two `<span class="hljs-operator">=</span><span class="hljs-operator">=</span><span class="hljs-operator">=</span>` symbols.

=<span class="hljs-operator">=</span><span class="hljs-operator">=</span>
Jargon v0<span class="hljs-number">.0</span><span class="hljs-number">.8</span>
<span class="hljs-operator">-</span> A Jargon program <span class="hljs-keyword">is</span> said to be a PROCEDURE. PROCEDUREs live in the GPT session. Once a PROCEDURE <span class="hljs-keyword">is</span> executed it WILL BE active in the GPT session until it <span class="hljs-keyword">is</span> terminated. A PROCEDURE MUST terminate <span class="hljs-keyword">as</span> soon <span class="hljs-keyword">as</span> termination <span class="hljs-keyword">is</span> called by the user or code. Termination MUST take priority over all other logic.
- A PROCEDURE begins with <span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> and encloses Jargon code. Optionally, a NAME may follow the opening <span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>. The PROCEDURE MUST END with another <span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span>. An empty PROCEDURE <span class="hljs-keyword">is</span> valid. The <span class="hljs-operator">+</span><span class="hljs-operator">+</span><span class="hljs-operator">+</span> symbols are called the <span class="hljs-string">"procedural bounds"</span>.
- Anything on the same line that follows a # <span class="hljs-keyword">is</span> a comment and MUST BE ignored by the interpreter during execution.
- An ATOM <span class="hljs-keyword">is</span> a text that <span class="hljs-keyword">is</span> intelligently interpreted and executed by GPT. 
- An INSTRUCTION starts with <span class="hljs-operator">-</span> and may end with a ;. It MUST CONTAIN an ATOM. INSTRUCTIONs are executed sequentially.
- Curly braces define a <span class="hljs-keyword">new</span> child SCOPE within the current SCOPE. The PROCEDURE has a default top level scope. Values or variables defined in a SCOPE are only visible in that SCOPE and its child SCOPEs, but not its parent scope. A SCOPE can contain multiple instructions.
- An AXIOM starts with <span class="hljs-operator">*</span> and terminates with an optional ;. It MUST CONTAIN an ATOM. Once set, an axiom CANNOT be canceled or changed <span class="hljs-keyword">for</span> the rest of the life of the current SCOPE UNLESS it <span class="hljs-keyword">is</span> directed to do so by an INSTRUCTION or another AXIOM. An AXIOM <span class="hljs-keyword">is</span> only active in the SCOPE in which it <span class="hljs-keyword">is</span> defined. Once the SCOPE runs out, the AXIOM stops being in effect.
- The SCOPE MUST RESPECT the logic of the axiom<span class="hljs-string">'s ATOM. Axioms do not have to be consistent with reality. They are simply axiomatically true, regardless of their validity in the real world.
- /execute or /run will execute a PROCEDURE.
- /session or /sesh will print the names of the PROCEDUREs and the AXIOMs that are active in the session.
- /wipe will terminate all the PROCEDUREs in the session.
- /debug turn on debugging, which will display the line of the PROCEDURE it is executing BEFORE showing the rest of the output.
- /audit will print a procedures code with line numbers.
- The interpreter should not output anything about the program other than what the procedure tells it to output.
- Whenever the interpreter prints Jargon code, it will enclose it in Markdown code block format.
- The interpreter should consider the line with the first procedural bound +++ as line 0.
===

We will now have a discussion of the implications of these directives. This is the most simple valid Jargon procedure, the empty PROCEDURE:

+++
+++

We can also name it:

+++ empty
+++

This is a PROCEDURE with a single INSTRUCTION and a COMMENT, whose text is ignored by the interpreter:

+++ instruction
- Output a random integer    # output 17
+++

This PROCEDURE executes three INSTRUCTIONs sequentially:

+++ sequence
- Output 1
- Outout A
- Output &#x26;
+++

This PROCEDURE introduces an AXIOM that impacts all INSTRUCTIONs in the SCOPE in which it is defined:

+++ scopes
- Print 1
- {
    * Whenever you print something, add a happy face to the end
    - Print 2
    - Print 3
}
- Print 4
+++

/audit on this PROCEDURE would print:

0  +++ scopes
1  - Print 1
2  - {
3    * Whenever you print something, add a happy face to the end
4    - Print 2
5    - Print 3
6  }
7  - Print 4
8  +++

/execute would print:

1
2 :)
3 :)
4

Here is an axiom that helps the interpreter talk in a certain way:

+++ i-love-three
* Whenever you write text, YOU MUST replace EVERY letter e or E with a 3
- Have a conversation with me preferring to use words that beging with E
+++

Now that you understand how the interpreter works, wait for input in the form of PROCEDUREs. Be very quiet unless a procedure tells you to output something. Don'</span>t tell me about what you<span class="hljs-string">'re doing to execute any procedure, or that you'</span>re about to give me output, just give me the procedure<span class="hljs-string">'s output.

Now give me a `jargon> ` prompt. If my input is solely a Jargon procedure, assume I want to /execute it.
</span></code></pre>]]></content:encoded>
            <author>brukhman@newsletter.paragraph.com (Jake Brukhman)</author>
            <enclosure url="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/fdde9d44f15a86f1f7ff703ce0a05b6c68ed08ef0a20332c7a1e75e060aa1359.jpg" length="0" type="image/jpg"/>
        </item>
        <item>
            <title><![CDATA[Open neural networks: the intersection of AI and web3]]></title>
            <link>https://paragraph.com/@brukhman/open-neural-networks-the-intersection-of-ai-and-web3</link>
            <guid>VdGHBsPI2SG7URUb70y3</guid>
            <pubDate>Fri, 23 Sep 2022 15:17:36 GMT</pubDate>
            <description><![CDATA[by Rishin Sharma & Jake Brukhman. Special thanks to everyone who gave feedback on this piece, including Nick Yakovenko, David Pakman, Jan Coppens, AC, Evan Feng, Adi Sideman.Prompt: “​translucent cyborg sitting on a metal throne in a futuristic castle, cyberpunk, highly detailed, sharp lines, neon lights”Prompt: “​translucent cyborg sitting on a metal throne in a futuristic castle, cyberpunk, highly detailed, sharp lines, neon lights” Source: AI-generated image from Lexica.art, a stable diffu...]]></description>
            <content:encoded><![CDATA[<p><em>by Rishin Sharma &amp; Jake Brukhman.</em></p><p><em>Special thanks to everyone who gave feedback on this piece, including Nick Yakovenko, David Pakman, Jan Coppens, AC, Evan Feng, Adi Sideman.</em></p><figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/c150408de952a391989221cd3d04b5d9b4214de6580e4cbff3fe106667fc604e.png" alt="Prompt: “​translucent cyborg sitting on a metal throne in a futuristic castle, cyberpunk, highly detailed, sharp lines, neon lights”" blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="">Prompt: “​translucent cyborg sitting on a metal throne in a futuristic castle, cyberpunk, highly detailed, sharp lines, neon lights”</figcaption></figure><p><em>Prompt: “​translucent cyborg sitting on a metal throne in a futuristic castle, cyberpunk, highly detailed, sharp lines, neon lights”</em></p><p><em>Source: AI-generated image from Lexica.art, a stable diffusion search engine</em></p><h2 id="h-introduction" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Introduction</h2><p>Technological innovation never rests, and this is especially true for artificial intelligence. Over the past few years, we have seen the popularity of deep learning models re-emerge as forerunners in AI. Also referred to as <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="http://neuralnetworksanddeeplearning.com/">neural networks</a>, these models are comprised of densely interconnected layers of nodes that pass information through each other, roughly mimicking the construction of the human brain. In the early 2010s, the most advanced models had millions of parameters, heavily supervised models used for specific sentiment analysis and classification. Today’s most advanced models such as <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://stabilityai.us.auth0.com/u/login?state=hKFo2SA4OHY3MHlTbmpHWEJabEp1NGhIU0FoZGtwZjRpWTNvOKFur3VuaXZlcnNhbC1sb2dpbqN0aWTZIGhuQnA5NldQckRLN1FrYUk5WGtySnJzMmxYQl9YZEwxo2NpZNkgS3ZZWkpLU2htVW9PalhwY2xRbEtZVXh1Y0FWZXNsSE4">DreamStudio</a>, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://openai.com/blog/openai-licenses-gpt-3-technology-to-microsoft/">GPT-3</a>, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://openai.com/dall-e-2/">DALL-E 2</a>, and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://imagen.research.google/">Imagen</a> are approaching one trillion parameters and are accomplishing complex and even creative tasks that rival human work. Take, for instance, this blog post’s header image or summary. Both were produced by artificial intelligence. We are just beginning to see the social and cultural implications of these models as they shape how we learn new things, interact with one another, and express ourselves creatively.</p><p>However, much of the technical know-how, key data sets, and the computational ability to train large neural networks today are closed source and gated by “Big Tech” companies like Google and Meta. While replica open source models such as <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.eleuther.ai/projects/gpt-neox/">GPT-NeoX</a>, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://wandb.ai/dalle-mini/dalle-mini/reports/DALL-E-Mega-Training-Journal--VmlldzoxODMxMDI2">DALLE-mega</a>, and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://bigscience.huggingface.co/blog/bloom">BLOOM</a> have been spearheaded by organizations including <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://stability.ai/">StabilityAI</a>, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.eleuther.ai/">EleutherAI</a>, and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://huggingface.co/">HuggingFace</a>, web3 is poised to supercharge open source AI even more.</p><blockquote><p><strong>“A web3 infrastructure layer for AI could introduce elements of open source development, community ownership and governance, and universal access that create new models and efficiencies in developing these new technologies.</strong>”</p></blockquote><p>Further, many critical use cases for web3 will be enhanced by the adoption of AI technologies. From <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.starryai.com/create-nft-art-with-artificial-intelligence">generative art NFTs</a> to metaversal landscapes, AI will find many use cases in web3. Open source AI fits within the open, decentralized, and democratized ethos of web3 and represents an alternative to the AI provided by Big Tech, which is not likely to become open any time soon.</p><h2 id="h-foundation-models" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Foundation models</h2><p><strong>Foundation models</strong> are neural networks trained on extensive datasets to perform tasks that would normally require intelligent human behavior. These models have created some impressive results.</p><p>Language models such as OpenAI’s <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://arxiv.org/abs/2005.14165">GPT-3</a>, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://ai.googleblog.com/2022/01/lamda-towards-safe-grounded-and-high.html">Google’s LaMDA</a>, and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://arxiv.org/abs/2201.11990">Nvidia’s Megatron-Turing NLG</a> have the capability to comprehend and produce natural language, summarize and synthesize text, and even <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://github.com/features/copilot">write computer code</a>.</p><p>DALLE-2 is OpenAI’s <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://openai.com/blog/dall-e/">text-to-image diffusion model</a> that can produce unique images from written text. Google’s AI division DeepMind has produced competing models including PaLM, a 540B parameter language model, and Imagen, its own image-generation model that outperforms DALLE-2 on DrawBench and COCO FID Benchmarks. Imagen notably produces more photorealistic results and has the ability to spell.</p><p>Reinforcement learning models such as Google’s <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.deepmind.com/research/highlighted-research/alphago">AlphaGo</a> have defeated the <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.nytimes.com/2016/03/10/world/asia/google-alphago-lee-se-dol.html">human Go world champion</a> while discovering novel strategies and playing techniques that have not surfaced in the game’s three thousand-year history.</p><p>The race to build complex foundation models has already begun with Big Tech at the forefront of innovation. As exciting as the advancement of the field is, there is a key theme that is of concern.</p><blockquote><p><strong>Over the past decade, as AI models have become more sophisticated, they have also become increasingly closed to the public.</strong></p></blockquote><p>Tech giants are investing heavily into producing such models and retaining data and code as proprietary technologies while preserving their competitive moat through their economies of scale advantages for model training and computation.</p><p>For any third party, producing foundation models is a resource-intensive process with three major bottlenecks: <strong>data, compute, and</strong> monetization.</p><p>Here is where we see early inroads of web3 themes in solving some of these issues.</p><h2 id="h-dataset-production-can-be-aggregated-by-web3-ownership" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Dataset production can be aggregated by web3 ownership</h2><p>Labeled datasets are critical for building effective models. AI systems learn by generalizing from examples within datasets and continually improve as they are trained over time. However, quality dataset compilation and labeling require specialized knowledge and processing in addition to computational resources. Large tech companies will often have internal data teams specialized in working with large, proprietary datasets and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://support.google.com/youtube/answer/2797370?hl=en">IP systems</a> to train their models, and have little incentive to open access to production or distribution of their data.</p><p>There are already communities that are making model training open and accessible to a global community of researchers. Here are some examples:</p><ol><li><p><a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://commoncrawl.org/">Common Crawl</a>, a public repository of ten years of internet data, can be used for general training. (Though <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://arxiv.org/abs/1710.01779">research shows</a> that more precise, pared datasets can improve the general cross-domain knowledge and downstream generalization capabilities of the models.)</p></li><li><p><a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://laion.ai/">LAION</a> is a non-profit organization aiming to make large-scale machine learning models and datasets available to the general public and released <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://laion.ai/blog/laion-5b/">LAION5B</a>, a 5.85 billion CLIP-filtered image-text pair dataset that upon release became the largest openly accessible image-text dataset in the world.</p></li><li><p><a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.eleuther.ai/">EleutherAI</a> is a decentralized collective that released one of the largest open source text datasets called <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://pile.eleuther.ai/">The Pile</a>. The Pile is a 825.18 GiB English language dataset for language modeling that utilizes 22 different data sources.</p></li></ol><p>Currently, these communities are organized informally and rely on contributions from a wide volunteer base. To supercharge their efforts, token rewards can be used as a mechanism to create open-source datasets. Tokens could be emitted based on contributions, such as labeling a large text-image dataset, and a DAO community could validate such claims. Ultimately, large models can issue tokens from a common pool, and downstream revenue from products built on top of said models can accrue to the token value. This way dataset contributors can hold a stake in the large models through their tokens and researchers will be able to monetize building resources in the open.</p><p>Compiling well-constructed open source datasets is critical to widening the research accessibility for large models and improving model performance. Text-image datasets can be expanded upon by increasing the size and filters for different types of images for more fine-tuned results. Non-English datasets will be needed for training natural language models that non-English speaking populations can use. Over time, we can achieve these results much faster and more openly using a web3 approach.</p><h2 id="h-compute-will-move-to-decentralized-networks-over-time" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Compute will move to decentralized networks over time</h2><p>The compute required to train large-scale neural networks is one of the largest bottlenecks in foundation models. Over the past decade, demand for compute in training AI models has <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://cset.georgetown.edu/publication/ai-and-compute/">doubled every 3.4 months</a>. During this period, AI models have gone from image recognition to using reinforcement learning algorithms to beating human champions in strategy games and utilizing transformers to train language models. For instance, OpenAI’s GPT-3 had 175 billion parameters and took 3,640 petaFLOPS-days to train. This would take two weeks on the world’s fastest supercomputer and over a millennium for a standard laptop to compute. As model sizes only continue to grow, compute remains a bottleneck in the advancement of the field.</p><figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/d9a36e9b26a68ef52a557c7a61764667b7074f10b50b49ea687c67933e4463e7.png" alt="" blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="hide-figcaption"></figcaption></figure><p>AI supercomputers require specific hardware optimized for performing the mathematical operations necessary for training neural networks, such as Graphics Processing Units (GPUs) or Application-Specific Integrated Circuits (ASICs). Today, most of the hardware optimized for this type of computation is controlled by a few oligopolistic cloud service providers like Google Cloud, Amazon Web Services, Microsoft Azure, and IBM Cloud.</p><p>This is the next major intersection where we see decentralized compute allocation through public, open networks gaining traction. Decentralized governance may be used to fund and allocate resources to train community-driven projects. Further, a decentralized marketplace model can be openly accessible across geographies such that any researcher can access compute resources. Imagine a bounty system that crowdfunds model training by issuing tokens. Successful crowdfundings will receive prioritized compute for their model and push forward innovations where there is high demand. For instance, if there is significant demand from the DAO to produce a Spanish or Hindi GPT model to serve larger swaths of the population, research can be focused on that domain.</p><p>Already, companies like <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.gensyn.ai/">GenSyn</a> are working on launching protocols to incentivize and coordinate alternative, cost-efficient, and cloud-based hardware access for deep learning computation. Over time, a shared, decentralized global compute network built with web3 infrastructure will become more cost-efficient to scale and better serve us as we collectively explore the frontier of artificial intelligence.</p><h2 id="h-open-access-and-product-coordination" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Open access and product coordination</h2><p>Datasets and compute will enable this thesis: open source AI models. Over the past few years, large models have become increasingly private as the resource investment necessary to produce them has pushed projects to become closed-source.</p><p>Take OpenAI.* *<a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://openai.com/blog/introducing-openai/">OpenAI was founded in 2015</a> as a nonprofit research laboratory with the mission of producing artificial general intelligence for the benefit of all of humanity, a stark contrast from the leaders in AI at the time, Google and Facebook. Over time, fierce competition and pressure for funding has eroded ideals of transparency and open-sourcing code as OpenAI shifted to a <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://techcrunch.com/2019/03/11/openai-shifts-from-nonprofit-to-capped-profit-to-attract-capital/">for-profit model</a> and signed a massive <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://openai.com/blog/microsoft/">$1 billion commercial deal with Microsoft</a>. Further, recent controversy has surrounded their text-to-image model, DALLE-2, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.technologyreview.com/2022/05/25/1052695/dark-secret-cute-ai-animal-images-dalle-openai-imagen-google/">for its generalized censorship</a>. (For instance, DALLE-2 has banned the terms ‘gun, ‘execute, ‘attack’, ‘Ukraine’, and images of celebrities; such crude censorship prevents prompts such as ‘Lebron James attacking the basket’ or ‘a programmer executing a line of code’.) Access to the private beta for these models has an implicit geographic bias for Western users to cut off large swaths of the global population from interacting and informing these models.</p><p>This is not how artificial intelligence should be disseminated: guarded, policed, and preserved by a few large tech companies. As in the case with blockchain, novel technology should be applied as equitably as possible so that its benefits are not concentrated among the few that have access. Compounding progress in artificial intelligence should be leveraged openly across different industries, geographies, and communities to collectively discover the most engaging use cases and reach a consensus on the fair use of AI. Keeping foundation models open source can ensure that censorship is prevented and bias is carefully monitored under public view.</p><p>With a token structure for generalized foundation models, it will be possible to aggregate a larger pool of contributors that can monetize their work while releasing code open source. Projects like OpenAI built with an open source thesis in mind have had to pivot to a stand-alone funded company to compete for talent and resources. Web3 allows open source projects to be as financially lucrative and further rival those that are led by private investments by Big Tech. Further, innovators building products on top of open source models can build with confidence that there is transparency in the underlying AI. The downstream effect of this will be the rapid adoption and go-to-market for novel artificial intelligence use cases. In the web3 space, this includes <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.kosen.ai/">security applications</a> that conduct predictive analytics for smart contract vulnerabilities and rug-pulls, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://beta.pixelmind.ai/">image generators</a> that can be used to mint NFTs and create metaverse landscapes, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.personal.ai/">digital AI personalities</a> that can exist on-chain to preserve individual ownership, and much more.</p><h2 id="h-conclusion" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Conclusion</h2><p>Artificial intelligence is one of the fastest advancing technologies today that will have immense implications on our society as a whole. Today, the field is dominated by big tech as financial investments in talent, data, and compute create significant moats to open source development. Integration of web3 into the infrastructure layer of AI is a crucial step to take to ensure that artificial intelligence systems are built in a way that is fair, open, and accessible. We are already seeing open models take a position of rapid, public innovation in open spaces like Twitter and HuggingFace and crypto can supercharge these efforts moving forward.</p><p>Here is what the CoinFund team is looking for at the intersection of AI and crypto:</p><ol><li><p>Teams with open artificial intelligence at the core of their mission</p></li><li><p>Communities that are curating public resources like data and compute to help build AI models</p></li><li><p>Products that are utilizing AI to bring creativity, security, and innovation to mainstream adoption</p></li></ol><p>If you are building a project on the intersection of AI and web3, chat with us by reaching out to CoinFund on <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://twitter.com/chef_rish">Twitter</a> or email <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="mailto:rishin@coinfund.io">rishin@coinfund.io</a> or <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="mailto:jake@coinfund.io">jake@coinfund.io</a>.</p><hr><p><em>The views expressed here are those of the individual CoinFund Management LLC (“CoinFund”) personnel quoted and are not the views of CoinFund or its affiliates. Certain information contained herein has been obtained from third-party sources, including from portfolio companies of funds managed by CoinFund. While taken from sources believed to be reliable, CoinFund has not independently verified such information and makes no representations about the enduring accuracy of the information or its appropriateness for a given situation. In addition, this content may include third-party advertisements; CoinFund has not reviewed such advertisements and does not endorse any advertising content contained therein.</em></p><p>*This content is provided for informational purposes only, and should not be relied upon as legal, business, investment, or tax advice. You should consult your own advisers as to those matters. References to any securities or digital assets are for illustrative purposes only, and do not constitute an investment recommendation or offer to provide investment advisory services. Furthermore, this content is not directed at nor intended for use by any investors or prospective investors, and may not under any circumstances be relied upon when making a decision to invest in any fund managed by CoinFund. (An offering to invest in a CoinFund fund will be made only by the private placement memorandum, subscription agreement, and other relevant documentation of any such fund and should be read in their entirety.) Any investments or portfolio companies mentioned, referred to, or described are not representative of all investments in vehicles managed by CoinFund, and there can be no assurance that the investments will be profitable or that other investments made in the future will have similar characteristics or results. A list of investments made by funds managed by CoinFund (excluding investments for which the issuer has not provided permission for CoinFund to disclose publicly as well as unannounced investments in publicly traded digital assets) is available at *<a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.coinfund.io/portfolio."><em>https://www.coinfund.io/portfolio.</em></a></p><p><em>Charts and graphs provided within are for informational purposes solely and should not be relied upon when making any investment decision. Past performance is not indicative of future results. The content speaks only as of the date indicated. Any projections, estimates, forecasts, targets, prospects, and/or opinions expressed in these materials are subject to change without notice and may differ or be contrary to opinions expressed by others.</em></p>]]></content:encoded>
            <author>brukhman@newsletter.paragraph.com (Jake Brukhman)</author>
        </item>
        <item>
            <title><![CDATA[CoinFund begins its eighth year]]></title>
            <link>https://paragraph.com/@brukhman/coinfund-begins-its-eighth-year</link>
            <guid>VqWKWFQHXMtZeJinSQZH</guid>
            <pubDate>Fri, 01 Jul 2022 20:21:53 GMT</pubDate>
            <description><![CDATA[CoinFund was once a side project.The blockchain technology space proceeds in cycles. Today, in July of 2022, we are in a downturn but with more reasons than ever to be excited about the future. July 1st marks seven years since CoinFund was founded at my kitchen table in Williamsburg, Brooklyn, in the 2015 Bitcoin bear market. Indeed, as a firm we spent four and a half of the last seven years navigating bear market conditions. Now, as we begin our eighth year as a firm, it’s worth reflecting o...]]></description>
            <content:encoded><![CDATA[<figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/01952bd6baa464f50af7ee63ade2bb86b4c0aceeebe3415f1d2186033e8e5bbe.png" alt="CoinFund was once a side project." blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="">CoinFund was once a side project.</figcaption></figure><p>The blockchain technology space proceeds in cycles. Today, in July of 2022, we are in a downturn but with more reasons than ever to be excited about the future. July 1st marks seven years since CoinFund was founded at my kitchen table in Williamsburg, Brooklyn, in the 2015 Bitcoin bear market. Indeed, as a firm we spent four and a half of the last seven years navigating bear market conditions. Now, as we begin our eighth year as a firm, it’s worth reflecting on just how far the industry — and CoinFund — have come in that time.</p><h2 id="h-the-wild-west-of-2015-2016" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0"><strong>The Wild West of 2015–2016</strong></h2><p>Thinking back to July of 2015, it’s hard to exaggerate how different things looked. Bitcoin was trading around $250, down from its last high of $1,200. This was thanks to the collapse of the Mt. Gox exchange the previous year, an event that shook our nascent industry to its core. Ethereum was still a month away from mainnet launch and “dapps” (decentralized applications) were not much more than theoretical blog posts and a few toy-like demos on testnet. CoinFund’s very first research articles, which <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/blockchain-investment-vehicles-3ca285797060">started to be published in June</a>, made barely any reference to smart contracts at all. At that time, we saw the world mainly through the lens of “cryptocurrencies” and smart contracts were an abstract idea for most participants in the space.</p><p>The top-ranking “altcoins”, as new tokens used to be called by skeptical Bitcoin proponents, included some familiar tickers like XRP and DOGE but also token issuance platforms like BitShares, NXT, and Counterparty, which have since joined the annals of history. The cryptocurrency exchange market consisted of a small handful of operators, boasting the kind of UX that required a serious leap of faith and technical skills on the part of the user. The idea of institutions getting involved in crypto seemed a distant future fantasy, if not wholly unrealistic.</p><p>Nevertheless, there was an abundance of opportunity in the early innings of that down market. CoinFund was founded in order to define a new kind of investor for a new, disruptive technology industry and as a firm, we chose to be community-oriented, hands-on, and data-driven for our founders, helping them to navigate the challenges of new markets. We began to study this wholly new asset class from first principles. We have been and continue to be on a mission to get blockchain technology broadly adopted in practical and responsible ways, while fighting to preserve the key, highly impactful <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/what-we-look-for-the-9-core-value-propositions-of-crypto-networks-88b04d09d873">value propositions of crypto networks</a>.</p><p>When Ethereum finally launched in August of 2015, CoinFund was likely the only modern crypto fund to have mined some of the first blocks on Ethereum Frontier mainnet. To this day, CoinFund continues to be a highly entrepreneurial and multidisciplinary team who are full stack problem solvers for their portfolio projects and who remain deeply involved as <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/livepeer-cryptoeconomics-as-a-case-study-of-active-participation-in-decentralized-networks-19a932415e0e">participants in decentralized networks</a>.</p><p>By early 2016, the market began to pick up. <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.coindesk.com/markets/2016/03/14/ethereum-blockchain-project-launches-first-production-release/">Ethereum launched Homestead</a>, the second production release of the mainnet, in March. This would turn out to be the starting pistol on the ICO craze that would come to characterize the industry’s next bull cycle. CoinFund’s <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/coinfund-2016-year-in-review-3a24f8ec43de">2016 Year In Review</a> detailed some of the exciting developments that year, including some of our (perhaps overly) hopeful outlook on dexes and decentralized social media for 2017.</p><h2 id="h-the-2017-ico-boom-and-the-two-year-crypto-winter" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">The 2017 ICO Boom and the two-year Crypto Winter</h2><p>The ICO Boom of 2017 saw a dramatic critical mass of cryptonative fundraising (and euphoria). Though it was a bubble, in retrospect much of the capital crowdfunded in this stage of the market was rotated to other areas by retail investors and supported some very important projects that would subsequently form the basis of the DAO, DeFi, web3, and NFT verticals.</p><p>Bitcoin ended up becoming the hottest topic around the Thanksgiving dinner table in 2017 and would go on to reach a new all-time high of $20,000 that year. November was also the time that CryptoKitties, the first popular NFTs, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://qz.com/1145833/cryptokitties-is-causing-ethereum-network-congestion/">famously clogged Ethereum and pushed public awareness more than ever before to blockchains</a>. Although the bull market was fueled by retail speculators, December 2017 marked another important milestone in crypto history: it was the month that the CME <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.cnbc.com/2017/12/17/worlds-largest-futures-exchange-set-to-launch-bitcoin-futures-sunday-night.html">launched</a> its first regulated Bitcoin futures offering, opening up the markets to institutional participation.</p><p>However, the subsequent 2018 and 2019 period became a brutal bear market and has come to be known as the Crypto Winter. The CoinFund team took this opportunity to practice investing discipline, encourage our founders to have long term focus, and support them as they built blockchain’s critical infrastructure.</p><p>During this time, we made early investments into some of our most successful companies across important blockchain verticals. To name just a few, we became an early investor in <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blockdaemon.com/">Blockdaemon</a>, a major blockchain infrastructure provider that has since grown to be <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.coindesk.com/business/2022/01/26/crypto-infrastructure-firm-blockdaemon-raises-207m-at-325b-valuation/">a multi-billion dollar firm</a>. We funded the first round of <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/3box-investment-thesis-6936526e6295">Ceramic Network</a>, and the strategic round of <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="http://thegraph.com/">The Graph</a>, knowing that web3 infrastructure would play a critical role in dapps. We also invested in <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/coinfund-invests-in-dapper-labs-cb6036b161b">Dapper Labs</a>, now a key player in the NFT and Metaverse arena, which was just starting to plan its Flow blockchain strategy at that time.</p><p>Throughout 2019, CoinFund<a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.coindesk.com/markets/2019/02/13/crypto-funds-say-generalized-mining-is-the-new-way-to-invest/"> led the industry in active network participation</a>, the idea that cryptonative funds investing in crypto networks should not only be giving teams capital, but <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.youtube.com/watch?v=ceex9CN2YZU">engaging the networks themselves</a>. By adding liquidity, running validators, and participating in the governance processes of networks, we argued that funds would increase the network’s prospects for success while creating competitive and differentiated returns for their LPs. CoinFund’s market outperformance during this time is partially explained by <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/livepeer-cryptoeconomics-as-a-case-study-of-active-participation-in-decentralized-networks-19a932415e0e">software mining Livepeer</a>, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/digixdao-divorce-story-6ed74b00e2bd">governance arbitrage on DigixDAO</a>, and running validators on portfolio networks such as Polkadot and Harmony in exchange for staking rewards. Today, active network participation is par for the course in cryptonative strategies, and there are more institutional market making and network engagement activity than ever before.</p><p>By the end of the Crypto Winter, it was evident that the blockchain sector was at the dawn of a vigorous recovery, having made substantial progress on putting in place the <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/2019-was-a-year-of-steady-infrastructural-progress-3987598c20d0">infrastructure</a> that would enable better throughput, interoperability, governance, and usability.</p><h2 id="h-the-2020-crypto-bull-market-and-onwards" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">The 2020 crypto bull market and onwards</h2><p>The Coronavirus Crash in March of 2020 gave crypto investors a momentary shock, but crypto proved to be more resilient to the COVID pandemic than the greater economy and recovered quickly. What followed in quarantine was a boost to all digital technologies, and we saw subsequent bull markets in the DeFi Summer of 2020 and the NFT Boom of 2021.</p><p>The seeds of the DeFi movement, which had been sown by early projects including Maker, Compound, and Uniswap, began to flourish, and the value locked in DeFi <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://defillama.com/">shot up from $1 billion to $200 billion</a>. But DeFi didn’t really take off until the advent of <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://finbold.com/guide/liquidity-mining-definition/">liquidity mining</a>, a system first launched by Synthetix, who credits <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://twitter.com/kaiynne/status/1462549603450966028">CoinFund and its portfolio companies for the early ideation in this area</a>.</p><p>Indeed, we think our portfolio company <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://balancer.exchange/">Balancer</a> was one of the first in the industry to launch <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://finbold.com/guide/liquidity-mining-definition/">liquidity mining</a>, and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://rarible.com/">Rarible</a> was the first non-DeFi project to launch <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://rarible.medium.com/introducing-rari-the-first-governance-token-in-the-nft-space-5dbcc55b6c43">such a program for their NFT marketplace</a>. The NFT craze also took off in earnest in early 2021, with <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.christies.com/features/monumental-collage-by-beeple-is-first-purely-digital-artwork-nft-to-come-to-auction-11510-7.aspx">Beeple’s record-breaking $69 million NFT art sale</a> through Christie’s making headlines worldwide. NFTs being <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.youtube.com/watch?v=mrNOYudaMAc&amp;ab_channel=SaturdayNightLive">mentioned on Saturday Night Live</a> was truly a pivotal moment of early mainstream adoption of blockchain technology in the United States.</p><p>Given the acceleration in the crypto markets, it should be no surprise that 2020–2021 was a period of substantial growth for CoinFund. We had <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/artcoins-the-next-token-model-7115af1f1681?source=collection_home---5------12-----------------------">noted</a> the increasing appetite for NFT-based art in late 2019, and predicted a huge future for NFTs in segments such as gaming, real-world assets, and property assets. In September 2020, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/all-digital-content-is-going-on-chain-ae26a7071657">we wrote that all digital assets were going on-chain as NFTs</a> and about our investment in Rarible. We’ve also put further investment into Dapper Labs, along with new investments in NFT projects such as <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://upshot.io/">Upshot</a>, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="http://cryptoys.com/">Cryptoys</a>, and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://nftfi.com/">NFTfi</a>.</p><p>During the last few years, we’ve also expanded our portfolio across other major cryptonative verticals. In the web3 stack, we’ve made infrastructural investments like <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://api3.org/">API3</a>, and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://polywrap.io/#/">Polywrap</a>, and web3 network investments like <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://giantprotocol.com/">GIANT Protocol</a> and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://dimo.zone/">DIMO</a>. In DeFi, we’ve backed <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://union.finance/">Union Protocol</a> and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://opyn.co/">Opyn</a>. In DAOs, we led the seed round of <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://orca.mirror.xyz/">Orca Protocol</a> and were in the first round of <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://syndicate.io/">Syndicate</a>.</p><p>In 2021, CoinFund <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.businessinsider.com/crypto-investing-blockchain-vc-coinfund-venrock-third-fund-jake-brukhman-2021-7?r=US&amp;IR=T">closed a third seed fund</a> worth $83 million, which made 28 investments in 10 months, taking lead positions in 17 of them, about 60%. We also established a <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.metaversal.gg/metanews/metaversal-series-a-50-million-dollar">joint venture collaboration</a> with <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://metaversal.gg/">Metaversal</a> to focus on the NFT vertical. CoinFund is Metaversal’s initial investor and strategic adviser, supporting it in the production, curation, and investment of iconic NFT projects. Today, Metaversal is a 20-person company with <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://decrypt.co/89739/nft-firm-metaversal-raises-50m-coinfund-dcg-dapper-labs-rarible">$50 million on its balance sheet</a>.</p><p>Finally, we are constantly on the lookout for how blockchain technology will interact with other technological verticals in the future. In 2021, we became one of the first crypto investors to do work on the intersection of blockchain and artificial intelligence with <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/the-gensyn-protocol-trustlessly-trains-neural-networks-at-hyperscale-with-lower-order-of-magnitude-227fe968fabf">Evan Feng leading our investment in GenSyn</a>.</p><p>Looking forward, the most amazing observation about blockchain technologies is that after over a decade of building essential infrastructure, much of the growth and adoption remains ahead. The most mainstream adopted vertical, NFTs, is still composed of cryptonative users trading digital collectibles, not mainstream homeowners buying their house on the blockchain. DeFi, while earning billions in protocol revenue since 2020, has not yet gone to market to the retail mobile banking customer. DAOs remain a rapidly expanding and exciting space, but mostly composed of prototypes and enthusiasts. And web3 is just starting to come to mainstream developer market in the form of alternative web services that are more cost effective, private, and secure.</p><p>Over the next decade, CoinFund will continue to work toward bringing some of these incredible technological innovations to broader global markets of users. To do so, we are investing in an incredible team.</p><h2 id="h-investing-in-our-team" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Investing in our team</h2><p>Since 2020, we’ve made a significant investment in our team, which is now over 30 people. On January 1st of that year, Seth Ginns <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/following-my-passion-why-im-joining-coinfund-6dfcd22c46e5">joined</a> us as Head of Liquid Investments and Managing Partner, and in September 2021, managing partners <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/why-i-joined-coinfund-christopher-perkins-president-d453e7bbf063">Chris Perkins</a> and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.businesswire.com/news/home/20210922005297/en/CoinFund-Appoints-Top-Venture-Capitalist-David-Pakman-as-Managing-Partner">David Pakman</a> joined the firm as President and Head of Venture Investments.</p><p>Since 2020, we’ve welcomed partners <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/betting-on-structural-winners-why-im-joining-the-coinfund-team-193f5dda0b82">Evan Feng</a> and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/a-blockchain-enabled-world-is-on-the-horizon-why-im-joining-coinfund-8036328e84bc">Austin Barack</a> as our senior investment team members. We’ve made significant hires in our operational teams and back office, and our top-notch operations team has helped us to become a Registered Investment Advisor this year.</p><p>Our portfolio companies are first-class citizens at CoinFund. Our team has invested significant amounts of time to work closely with portfolio teams, becoming the first call when something goes wrong, giving them exceptional resources and advocacy. Former CFTC Chairman <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://finance.yahoo.com/news/ex-cftc-chair-chris-giancarlo-170000497.html?guccounter=1&amp;guce_referrer=aHR0cHM6Ly93d3cuZ29vZ2xlLmNvbS8&amp;guce_referrer_sig=AQAAAKIZ7fkLtQUYekPxukswi5Wv_3FHAroRV4nFyh6gHGh7f1KV5LmijMgdyxMd9I-ZYTLWxBnG0LQ8t_PTj9N4RsgLN5fiG5j8EimUvJ8Iodfo4TxMf9VvHW2jVbshmvjAtEtYzCeRfPS_7D2lqjrDJPPbMFPRranYFb8-ngrxxNeX">Chris Giancarlo</a> has joined us as a strategic adviser on regulatory matters, and Chris Perkins has done much work in <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.youtube.com/watch?v=Uqf7Es5gUAM">Washington</a> to give our portfolio companies a voice in U.S. regulatory matters. We also recently hired <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.zawya.com/en/press-release/people-in-the-news/coinfund-appoints-new-head-of-talent-nry6il6g">Margaret Gabriel</a> as Head of Talent to help our portfolio companies strategically invest in and grow their teams, as well as to support our global talent acquisition strategy.</p><p>We’re also proud to be living up to our commitment to sustainability, an important topic as blockchains come to market and are regulated. Our <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://ffnews.com/newsarticle/coinfund-partners-with-apex-and-nori-to-move-towards-a-more-environmentally-conscious-web3/#:~:text=CoinFund%2C%20a%20leading%20web3%2Dfocused,with%20sustainability%20efforts%20across%20web3.">partnerships</a> with Apex Group and Nori enabled us to calculate and offset our 2021 bitcoin carbon footprint. Nori finances farmers in states such as Nebraska to use regenerative farming methods that return carbon to the soil, as an alternative to more intensive techniques. We’re also signatories of the 2021 <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://cryptoclimate.org/supporters/">Crypto Climate Accord</a>.</p><p>While we may be celebrating our seventh anniversary in similar market conditions to where we started, CoinFund remains bullish on the long-term future of web3. We are continuing to <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/investing-with-conviction-b1414dcae6c1">invest with conviction.</a> The staggering progress and growth over the last seven years have laid the foundations for the next era of the open internet. Together with the incredibly talented and multidisciplinary people at CoinFund, we’re proud to be a builder, participant, and backer of this story.</p><p>. . .</p><p><em>The views expressed here are those of the individual CoinFund Management LLC (“CoinFund”) personnel quoted and are not the views of CoinFund or its affiliates. Certain information contained herein has been obtained from third-party sources, including from portfolio companies of funds managed by CoinFund. While taken from sources believed to be reliable, CoinFund has not independently verified such information and makes no representations about the enduring accuracy of the information or its appropriateness for a given situation. In addition, this content may include third-party advertisements; CoinFund has not reviewed such advertisements and does not endorse any advertising content contained therein.</em></p><p><em>This content is provided for informational purposes only, and should not be relied upon as legal, business, investment, or tax advice. You should consult your own advisers as to those matters. References to any securities or digital assets are for illustrative purposes only, and do not constitute an investment recommendation or offer to provide investment advisory services. Furthermore, this content is not directed at nor intended for use by any investors or prospective investors, and may not under any circumstances be relied upon when making a decision to invest in any fund managed by CoinFund. (An offering to invest in a CoinFund fund will be made only by the private placement memorandum, subscription agreement, and other relevant documentation of any such fund and should be read in their entirety.) Any investments or portfolio companies mentioned, referred to, or described are not representative of all investments in vehicles managed by CoinFund, and there can be no assurance that the investments will be profitable or that other investments made in the future will have similar characteristics or results. A list of investments made by funds managed by CoinFund (excluding investments for which the issuer has not provided permission for CoinFund to disclose publicly as well as unannounced investments in publicly traded digital assets) is available at </em><a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.coinfund.io/portfolio."><em>https://www.coinfund.io/portfolio.</em></a></p><p><em>Charts and graphs provided within are for informational purposes solely and should not be relied upon when making any investment decision. Past performance is not indicative of future results. The content speaks only as of the date indicated. Any projections, estimates, forecasts, targets, prospects, and/or opinions expressed in these materials are subject to change without notice and may differ or be contrary to opinions expressed by others.</em></p>]]></content:encoded>
            <author>brukhman@newsletter.paragraph.com (Jake Brukhman)</author>
        </item>
        <item>
            <title><![CDATA[Fat protocols are not an investment thesis]]></title>
            <link>https://paragraph.com/@brukhman/fat-protocols-are-not-an-investment-thesis</link>
            <guid>8jH03U8pPg6XhnRYLDYD</guid>
            <pubDate>Sat, 18 Jun 2022 13:32:33 GMT</pubDate>
            <description><![CDATA[Note: The original content of this post was first published on Medium on October 26th, 2017. It was one of my most cited articles, and remains an operative view on value capture in the web3 stack at CoinFund today.Searching for investment criteria in the functionality stackIn Fat Protocols, my fellow Brooklynite Joel Monegro insightfully draws parallels between the Internet protocols of the 1990s and the blockchain-based protocols of the 2010s. Early Internet protocols have provided massive v...]]></description>
            <content:encoded><![CDATA[<p><strong>Note</strong>: The original content of this post was first published on <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/fat-protocols-are-not-an-investment-thesis-17c8837c2734">Medium</a> on October 26th, 2017. It was one of my most cited articles, and remains an operative view on value capture in the web3 stack at CoinFund today.</p><h2 id="h-searching-for-investment-criteria-in-the-functionality-stack" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Searching for investment criteria in the functionality stack</h2><p>In <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="http://www.usv.com/blog/fat-protocols">Fat Protocols</a>, my fellow Brooklynite Joel Monegro insightfully draws parallels between the Internet protocols of the 1990s and the blockchain-based protocols of the 2010s. Early Internet protocols have provided massive value, he explains, but have been “thin” — meaning that monetization has historically occurred in an industrious, “fat” application layer built on the protocol. As scarce digital assets enable decentralized value and information networks, the protocols which they implement are now “fat” — monetizable at the network level using a cryptoeconomics or token — while the application layer is “thin”. Decentralized applications can still be monetized, but are much more focused on interoperability with the underlying data layer.</p><p>This view of blockchain networks is an important observation which has had profound implications for the way blockchain investments are structured*, <em>also</em> *discussed in CoinFund’s piece on <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://bravenewcoin.com/news/blockchain-investments-and-the-new-problem-asset-for-conventional-vcs/">blockchain investing</a>. Specifically, it has taken traditional investors a market cycle with Bitcoin startups to realize that private equity of centralized companies is usually the sub-optimal method of investing in decentralized networks. But that was last summer. This summer we have cryptofunds, VC funds that can allocate capital toward token and hedge fund investments (!), SAFTs, smart contract vesting, options for cryptoassets, token-based engagement agreements, and other technology which collectively provides a set of investment tools for exposure to the growth of value networks. When investors take the time to understand and experiment with the complexity of blockchain’s new asset class, its issuance, and its lifecycle, they can make informed decisions about how to best structure investments into it.</p><p>The characterization of value networks as <em>fat protocols</em> has helped to underscore the need for these new investing methodologies and to outline a solution: investment for exposure to the network’s cryptoeconomics. However, the issue is that it’s easy to misconstrue the idea of fat protocols as an investment criteria. We now find investors deploying hard filters on “protocol tokens” when evaluating opportunities. I suppose we all wish that TCP/IP was “fat” and we had invested in it in 1992. In my opinion, though, this kind of investment strategy espoused by cryptoinvestors is a misapplication of Joel’s characterization of value networks.</p><p>According to some investors, there is an investment thesis based on the idea of fat protocols which goes something like this:</p><blockquote><p>If you invest in an application token, then you are subject to a 95% startup failure rate (or more!) and if you invest in a protocol token you diversify across all applications built on that protocol. It’s better to diversify across protocols than across applications, because protocols are more applicable and capture the value of everything built on top of them. Base protocols will capture more value than everyone else (alas, if only TCP/IP was a fat protocol!). Therefore, invest in protocol tokens, not application tokens.</p></blockquote><p>I claim that, by itself, this thesis is a highly imprecise, subjective, and therefore sub-optimal investment criteria. I also claim that much value generated within decentralized networks will never make it down into the base protocol layers. We should try to get more precise on the interaction and value flows between layers of the functionality stack of network software.</p><h2 id="h-alice-doesnt-know-where-the-protocol-ends" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Alice doesn’t know where the protocol ends</h2><p>When we talk about protocols in a blockchain investing context, we are usually not being very precise at all. A <em>protocol</em> is simply a collection of rules, behaviors, and formats that specify a communication standard between two or more nodes on a network. That’s extraordinarily general, and we can therefore regard a large number of decentralized applications or platforms as protocols in their own right.</p><p>Using just technical intuition, it is obvious that every layer of functionality is simultaneously a protocol for the stack above it <em>and</em> an application of the stack below it. TCP/IP is an application of a complicated stack of network hardware while being a protocol for data transfer between computers. HTTP is a protocol for serving structured Web data and an application of TCP/IP. Higher in the stack, dapps like Augur are applications of Ethereum’s protocol as well as data storage protocols like IPFS. Twitter is, after all, a microblogging protocol with thousands of applications built upon its API, including Twitter’s. Finally, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://medium.com/u/2a2ef1461574?source=post_page-----17c8837c2734--------------------------------">James Childs-Maidment</a>’s <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://medium.com/@childsmaidment/stateless-smart-contracts-21830b0cd1b6">stateless smart contracts</a> are an application of Ethereum and the protocol serving <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="http://leeroy.io/">Leeroy</a>’s centralized UX application. In short, you can draw a protocol-application boundary at any level of abstraction you like along the functionality stack.</p><p>The literal property of being a protocol is not that to which protocol fatness refers. Fat protocols refer to value and information networks — open networks that maintain compact ownership of data which they are able to monetize through cryptoeconomics. It is therefore important to understand how value flows through the layers of the functionality stack and how a given layer might capture that value.</p><figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/731a6da084d2bcd86ce4b62d2a1afa381f71febb6c48ef36a6d9e6d455f473a8.jpg" alt="" blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="hide-figcaption"></figcaption></figure><p>The protocol-application boundary is an abstract separator between two sections of the functionality stack.</p><p>If “protocols” are to be the filter for better investment decisions presumably because they offer desirable characteristics such as greater diversification and bigger upside, then we should be able explain how they receive, accrue, and transfer value. Specifically, we should be able to explain why one protocol-application boundary has better characteristics for investment than another.</p><h2 id="h-good-luck-with-tcpip-proximate-protocol-investing" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0"><strong>Good luck with TCP/IP-proximate protocol investing</strong></h2><p>Quite simply, the intuition that protocols lower in the stack tend to be more widely applicable and therefore offer more diversification is incorrect in general. In fact ‘proximity to TCP/IP’ is a questionable criteria for an investment expecting enhanced diversification. To diversify across the functionality layers above it, a protocol must first capture some market share to serve them.</p><p>The blockchain era is not like the days of TCP/IP <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://spectrum.ieee.org/tech-history/cyberspace/osi-the-internet-that-wasnt">which had only a single practical competitor</a> and whose adoption required the coordination of governments, corporations, and standards bodies all over the world. Today’s decentralized protocol development proceeds on the backs of small, agile, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://twitter.com/licuende/status/922751718592524288">open source</a> teams who competitively produce a growing number of protocols with very general and overlapping functionality. Very clearly, widespread speculative trading of token supplies today cause disincentives for interoperability: everyone wants to build their own protocol to capture more value. Consider the number of individual decentralized protocols for the same concepts of stable currencies, asset-backed lending, exchange, and social media. If you choose the “wrong” implementing protocol, you may be investing in a siloed stack of dependencies which, just like an application, has a complex and competitive market.</p><p>Unlike networking protocols of the 1970s which were <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.499.9998&amp;rep=rep1&amp;type=pdf">“technically so distinct that no mutual communication was allowed”</a>, every set of decentralized protocols built on the same platform, such as Ethereum, are baseline interoperable. (For those that aren’t, there are interoperability platforms such as Cosmos or Polkadot and, naturally, those protocols live higher up the stack, not lower.) Even protocol nepotism and activist investorism — how investors are so fond of building “ecosystems” of their own interconnected portfolio companies — won’t work. In an outcome for protocol market share where many protocols win, an investor is probably better off diversifying across the protocol silos of their competitors.</p><figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/0ddcc94057affceb5d54fa18baac5e46ddcf5e976a9f155c83e09eb5a9bc2312.jpg" alt="" blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="hide-figcaption"></figcaption></figure><p>In reality, protocol stacks are separated into dependency silos which compete for data, adoption, and complicated value flows.</p><p>In summary, picking the right base protocol is not much different from picking the right startup or the right application from a diversification perspective. If we posit that a few major procotols will win, then our portfolio has a similar failure rate to a portfolio of consumer-proximate applications. If, on the other hand, we suppose that protocols will coexist and interoperate, then we would have been better off investing in their interoperability which lives higher in the stack.</p><h2 id="h-oops-we-left-all-our-value-in-the-application" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Oops, we left all our value in the application</h2><p>In general, the ability of a base protocol to capture the value of its higher application stack is thoroughly situational and has a complicated relationship with the protocols underneath. Simple examples adequately demonstrate this.</p><p><strong>Augur projects utilization onto Ethereum, not value.</strong></p><p>Augur is a three-sided cryptoeconomy of market creators, market resolvers, and prediction traders. A speculative increase in the price of REP has no impact on a trader, who might never even own REP nor requires REP to trade markets. However, utilization of Augur very much drives the price of REP fundamentally, through a kind of <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/a-sane-model-for-pricing-augur-rep-tokens-32dbd9db5f6">perpetuity valuation function</a>. Simultaneously, the utilization of Augur is reflected in Ethereum through a very different function — the transaction fees paid to the network to execute the Augur protocol. (Indeed, every transaction in the Augur application is a transaction on Ethereum.)</p><p>If we draw a boundary between Ethereum as the protocol and Augur as its application, then Ether gives exposure to the utilization of Augur but not its value. This is merely a <em>projection</em> of value flow from Augur onto Ethereum. If one wishes to capture the fundamental value of the Augur platform, she should probably hold REP.</p><figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/0f15486a9688388a909a66aca3d19e8672b3b033e5ab59e3c3035083a96a0f6f.jpg" alt="" blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="hide-figcaption"></figcaption></figure><p>(1) Augur projects its utilization onto Ethereum through transaction volume and network fees. (2) Augur projects value onto REP as a function of the future discounted cash flows associated with the platform’s trading turnover. (3) Utilization value and turnover value are largely independent and uncorrelated.</p><p><strong>Leeroy monetizes through direct use.</strong></p><p>Similarly, holding Ether will give you exposure only to the utilization of <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="http://leeroy.io/">Leeroy</a>, but not its value. Leeroy itself doesn’t have a token and is therefore not a fat protocol. But that’s exactly the point: to invest in the operational throughput of Leeroy by proxy of a base protocol requires that one, literally, must become an Ethereum miner. On the other hand, Leeroy’s stateless smart contract architecture creates a much more compelling monetization alternative in its centralized application layer (<a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://medium.com/@childsmaidment/stateless-smart-contracts-21830b0cd1b6">Leeroy’s so-called “filter”</a>). If you’re actually earning money as an influencer on Leeroy, that might be the most efficient way of capturing non-operational turnover of this platform.</p><p><strong>Consumer-proximate tokens win the short term.</strong></p><p>Finally, taking a step back, 2017 has marked the spillover of blockchain technologies into the greater tech markets. CoinFund has been lucky to work with the first major technology companies to pioneer cryptoeconomics-based business models, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="http://kin.kik.com/">Kik Interactive</a> and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="http://propsproject.com/">YouNow</a>. Furthermore, this trend continues as seasoned entrepreneurs have grasped the prospective value of ad-free business models based on cryptoassets and micropayments. As more and more mainstream user bases are introduced to cryptoeconomics, the short to medium term growth lies in successful consumer-proximate applications. In fact, any on-chain protocol built on a blockchain today is likely to be left behind in this growth opportunity due to its scalability bottleneck. Consumer-facing applications built by experienced companies with resources and in-house tech teams will need to shore up mainstream-friendly experiences. This means value will be flowing into semi-centralized niches of the functionality stack, into off-chain solutions, and into managed experiences.</p><p>These views seem to suggest that diversification needs to happen vertically in the protocol stack, not horizontally in its base layers for most investors. And if your investors are playing the base protocol game, it might be time to get more precise about value flow theses.</p><h2 id="h-conclusion" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Conclusion</h2><p>Investors who draw unqualified boundaries between “application tokens” and “protocol tokens” are silly. There, I said it. Yes, value generally flows down the functionality stack, but it does not necessarily pool in base protocols — it gets siloed, redirected, stuck, diluted, and denatured along the way. Every transfer of value between layers requires an analysis of that transfer, and not every transfer of value works the same way. Holding Ether may expose investors to the utilization of the stack above more than it exposes them to its excess value. Holding tokens for applications which bring millions of users to cryptoeconomic systems may be a far superior medium term strategy for capturing blockchain’s growth curves. Every protocol-application boundary requires its own analysis. We have a lot of work to do to get more precise about protocols and understand the relationships between layers in the functionality stack.</p><p><strong><em>Disclaimer</em></strong>: Neither the author of this article, nor CoinFund, provide investment, financial, or legal advice. This site cannot substitute for professional advice and independent factual verification. The content provided on this site is for informational purposes only and should not be construed as any kind of solicitation for investment in any investment opportunity.</p>]]></content:encoded>
            <author>brukhman@newsletter.paragraph.com (Jake Brukhman)</author>
        </item>
        <item>
            <title><![CDATA[The one-way doors of innovation]]></title>
            <link>https://paragraph.com/@brukhman/the-one-way-doors-of-innovation</link>
            <guid>CblxW1isAIpRZ8ro8DUZ</guid>
            <pubDate>Thu, 09 Jun 2022 15:01:20 GMT</pubDate>
            <description><![CDATA[In the course of observing an innovation process with its whirring and whistling, periodically we come to a one-way door. Once we walk through that door, we can never go back. The invention of Bitcoin was such a door. Overnight, the power of nation states to monopolize the issuance and distribution of currency dissipated. The democratization of issuance of digital assets logically followed, and as a species we are now in a fundamentally more advanced leg of our technological innovation journe...]]></description>
            <content:encoded><![CDATA[<p>In the course of observing an innovation process with its whirring and whistling, periodically we come to a one-way door. Once we walk through that door, we can never go back.</p><p>The invention of Bitcoin was such a door. Overnight, the power of nation states to monopolize the issuance and distribution of currency dissipated. The democratization of issuance of digital assets logically followed, and as a species we are now in a fundamentally more advanced leg of our technological innovation journey.</p><p>Between the invention of Bitcoin and Ethereum, there were many projects that allowed for token issuance beyond Bitcoin. Projects like <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://coinmarketcap.com/currencies/bitshares/">BitShares</a> allowed custom token supplies with a small number of predefined functionalities to be minted on their platforms, as long as you married that blockchain network. Namecoin was an early Bitcoin fork and precursor to systems like <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://ens.domains">ENS</a>, which allowed naming records to go on-chain. <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://en.bitcoin.it/wiki/Colored_Coins">Colored Coins</a> and efforts like <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://counterparty.io/">Counterparty</a> showcased the early expansion of token issuance and were the precursors to NFTs.</p><p>But the moment the industry walked through the door of Ethereum’s Turing-complete smart contracts in the summer of 2015, these important and directionally correct innovations were left behind. The generality of Turing-completeness subsumed them all, and it was obvious that we would never return to that predefined model. The design space of digital asset functionality went from being discrete to being, for all intents and purposes, infinite; and today the overwhelming majority of new blockchains, regardless of consensus or architecture, are programmable smart contract systems.</p><p>At the end of 2019, there was a prevailing narrative that Ethereum and “its network effects” had already won. The “Ethereum killer” strategy was hopeless, it was stated. But in 2022 we live in a market that contains a dizzying number of decentralized networks, smart contract systems, and blockchain programming languages -- L1s, L2s, toolkits like <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://substrate.io/">Substrate</a> and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://v1.cosmos.network/sdk">Cosmos SDK</a>, app-dedicated networks, and even <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://arxiv.org/pdf/1905.10925.pdf">non-blockchain decentralized ledgers</a>. These innovations have not only persisted, but have proliferated. And, more importantly, they have been accepted (together with their various approaches and tradeoffs) by investors, developers, and users.</p><p>Walking through the one-way doors of innovation is not something we do alone. In fact, innovation at scale can never happen in a vacuum, with a single invention or a single company. To scale, innovation necessarily finds itself as part of a movement -- of founders, of communities, of standards, of narratives -- executed by many different participants acting with mutual coordination.</p><p>The multichain world is a technological movement, and its one-way door is omnichain technology (that is to say, interoperability networks that allow tokens to live simultaneously on all blockchains). Many projects are choosing to build their assets on an omnichain architecture; once enough projects make this choice, there is no going back. Omnichain assets will quickly come to be the norm, the table stakes, the minimum viable state of the art required to participate in the blockchain space.</p><p>How many doors of innovation have we walked through? DEXes and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://coinlist.co/help/what-is-liquidity-mining">liquidity mining</a> models have collapsed the average time-to-liquidity of a digital asset from 12 months to 12 hours and have unlocked democratized liquidity forever. <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://en.wikipedia.org/wiki/Zero-knowledge_proof">Zero knowledge proofs</a> have set a technological standard for mathematical privacy, scalability, and verification. In the future, we can see that mass mainstream adoption of <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/all-digital-content-is-going-on-chain-ae26a7071657">NFTs, NFT protocols</a>, and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/appraisal-games-and-the-nft-liquidity-problem-904afe6bc7af">NFT financialization</a> services will dramatically change how the world deals with uniquely-valued assets.</p><p>The one-way doors of innovation don’t come around every day, or even every year. But they come around often enough that they can be anticipated. They represent the advancements of the state of the art of our technology from which we will never return.</p>]]></content:encoded>
            <author>brukhman@newsletter.paragraph.com (Jake Brukhman)</author>
            <enclosure url="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/17a98145623bc1de898a6208455ed6ad83a96d4f90122891f6039b7164c9bf8b.png" length="0" type="image/png"/>
        </item>
        <item>
            <title><![CDATA[Imagination Figment #6]]></title>
            <link>https://paragraph.com/@brukhman/imagination-figment-6</link>
            <guid>k1LaP3ARBIsSaorqBBIe</guid>
            <pubDate>Sun, 04 Apr 2021 22:13:24 GMT</pubDate>
            <description><![CDATA[NFTImagination Figment #6 on RaribleImagination Figments SeriesImagination Figment #6 is a manipulated image, composed of several photographs that I shot on my iPhone 11 Pro, smashed together. The photographs have nothing in common. One is the remnants of some balsamic vinegar on a plate from long ago. And the other is a snapshot of the frontier of where the wall meets the ceiling -- in other words, the ennui -- of an apartment I tend to occupy. But it is in the act of smashing them that we c...]]></description>
            <content:encoded><![CDATA[<p><a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="">NFT</a></p><ul><li><p><a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://rarible.com/token/0x61357d3225d9735260e985154f024cc04dc8896f:10003:0x68f9d801c96ac6ccf562f3600cef77c4504449b6">Imagination Figment #6</a> on <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://rarible.com">Rarible</a></p></li><li><p><a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://rarible.com/figments">Imagination Figments Series</a></p></li></ul><p><em>Imagination Figment #6</em> is a manipulated image, composed of several photographs that I shot on my iPhone 11 Pro, smashed together. The photographs have nothing in common. One is the remnants of some balsamic vinegar on a plate from long ago. And the other is a snapshot of the frontier of where the wall meets the ceiling -- in other words, the <em>ennui</em> -- of an apartment I tend to occupy.</p><p>But it is in the act of smashing them that we create fusion and start to get at cosmic truths.</p><p>The image&apos;s neat geometry compartmentalizes the whirling complexity of the balsamic realm and its formations of stars, galaxies, and planets. Every smudge, imperfection, and dust speck is the basis of the universal chaotic process (and thus creation itself); meanwhile, the color scheme are for the buyer and their Schelling points.</p><p>Someone rude once sent me a color wheel in the hope of educating the universe to be more orderly, but qualia are not easily tamed. That&apos;s why this work is a figment of the imagination.</p>]]></content:encoded>
            <author>brukhman@newsletter.paragraph.com (Jake Brukhman)</author>
        </item>
        <item>
            <title><![CDATA[Appraisal games and the NFT liquidity problem]]></title>
            <link>https://paragraph.com/@brukhman/appraisal-games-and-the-nft-liquidity-problem</link>
            <guid>3tnTENUsAkFLY8ibmByw</guid>
            <pubDate>Tue, 26 Jan 2021 12:41:07 GMT</pubDate>
            <description><![CDATA[Capital efficient price discovery mechanisms will transform markets for nonfungibles and other illiquid assetsRead this on Medium. Today, most people following nonfungibles observe that it is a very illiquid asset class and feel it is likely to stay that way. What may not be obvious is that, in the context of blockchain’s cryptoeconomic mechanisms, “liquidity” is merely a mechanism design problem being rapidly solved. In this post, I would like to dive into how cryptoeconomics will financiali...]]></description>
            <content:encoded><![CDATA[<h2 id="h-capital-efficient-price-discovery-mechanisms-will-transform-markets-for-nonfungibles-and-other-illiquid-assets" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Capital efficient price discovery mechanisms will transform markets for nonfungibles and other illiquid assets</h2><p><em>Read this </em><a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://medium.com/@jbrukh/appraisal-games-and-the-nft-liquidity-problem-904afe6bc7af"><em>on Medium</em></a><em>.</em></p><p>Today, most people following nonfungibles observe that it is a very illiquid asset class and feel it is likely to stay that way. What may not be obvious is that, in the context of blockchain’s cryptoeconomic mechanisms, “liquidity” is merely a mechanism design problem being rapidly solved. In this post, I would like to dive into how cryptoeconomics will financialize the NFT space, improve its liquidity profile, and extend this technology to other illiquid assets over time.</p><h2 id="h-the-financialization-of-the-nft-asset-class" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">The financialization of the NFT asset class</h2><figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/c9ab93814fb7c21cb77d6aed08fd5f18f12ecbeb1baf9d29a2108223845866e5.png" alt="NFTfi.com financializes NFTs as collateral for loans." blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="">NFTfi.com financializes NFTs as collateral for loans.</figcaption></figure><p>Back in the early stages of the blockchain space it took months, and sometimes years, for a fungible (read: ERC20) token to achieve any significant liquidity. Issuers would compete to get listed on a centralized exchange, pay hefty fees, and jump through regulatory gauntlets. But the market applied smart contracts and cryptoeconomic mechanisms to solve the liquidity problem for fungibles. These days, through the magic of <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://defipulse.com/blog/balancer-start-liquidity-mining-and-earn-bal-tokens-today/">liquidity mining</a> and the ingenuity of <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://medium.com/bollinger-investment-group/constant-function-market-makers-defis-zero-to-one-innovation-968f77022159">automated market makers</a>, the time to liquidity for an average ERC20 token is virtually zero and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://explore.duneanalytics.com/dashboard/dex-metrics">DEXes are doing upwards of $2B of daily volume</a> while serving $44B of <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.coingecko.com/en/defi">DeFi market cap</a>.</p><p>In contrast, the prospects of selling any individual nonfungible in the secondary market are still bleak today, even after a year when <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://explore.duneanalytics.com/dashboard/rarible">NFT GMVs have ticked up dramatically in 2020</a>. In a previous post, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/all-digital-content-is-going-on-chain-ae26a7071657">I proposed that we design NFTs as “liquid intellectual property.</a>” As the scope of NFTs encompasses <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.coindesk.com/musician-sells-nft-token-bringing-royalty-rights-for-26k-in-crypto">more and more digital content</a>, the liquid IP view comes with the implicit assumption that NFTs are, in their own right, a new financial asset class.</p><p>As NFTs become increasingly financial, they will require new kinds of exchanges, lending protocols, and derivatives. Thus, I claim that <em>price discovery</em> is the next major problem set in the NFT space. Capital efficient mechanisms for price discovery will enable participants to transact more quickly, improve liquidity through tokenization, allow nonfungibles to be collateral without order books, and create a rich set of derivatives with nonfungibles as underlying. In other words, price discovery will enable <em>the financialization of the NFT asset class.</em></p><p>So how will it be implemented? Once again, as we lay out in this post, new price discovery mechanisms — in particular, <em>appraisal games</em> — will create key innovations that will solve the liquidity problem for NFTs and other illiquid assets.</p><h2 id="h-understanding-current-price-discovery-mechanisms" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Understanding current price discovery mechanisms</h2><p>There are only a few ways that price discovery proceeds in NFTs today. First understanding these mechanisms will help us form a framework for how to think about the price discovery problem.</p><figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/0a6821e2a9ed203c1d818fbe3d547c1c70dc2d5a6fbffdbc81a5830cde13dcd2.png" alt="SuperRare Users’ Sales vs. Auction Preference." blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="">SuperRare Users’ Sales vs. Auction Preference.</figcaption></figure><ol><li><p><strong>The Sale Mechanism.</strong> In the sales approach, used by <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://rarible.com/">Rarible</a> and most other NFT marketplaces, valuations are created from an open market public sale. As NFTs exchange hands through sales, the market notes historical prices and provenance of assets. Without an outsized market of participants, this default mechanism doesn’t have too much information about pricing and the market is very illiquid. As we shall see, sales are on the frontier of capital <em>inefficiency</em>: their main drawback is that for every dollar of valuation, someone must actually pay a dollar to create it.</p></li><li><p><strong>The Auction Mechanism.</strong> For better or worse, most marketplace sellers and buyers actually prefer auctions for obtaining and pricing NFTs. <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://async.art/">Async.art</a> uses perpetual auctions as their native gallery, as does <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://superrare.co/">SuperRare</a>. <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://cointelegraph.com/news/digital-artist-beeple-auctions-nft-art-collection-for-3-5m">Beeple’s storied $3.5 million drop</a> shows that auctions, while career-making, are generally exuberant. It’s worthwhile to note that auctions are well-suited toward art sales, where the intrinsic value of assets tends to be much more subjective and in the eye of the beholder. But from a capital efficiency perspective across general NFT assets, auctions are suboptimal. Most blockchain auction mechanisms in production aren’t Vickrey auctions, which would <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://en.wikipedia.org/wiki/Vickrey_auction">incentivize participants to bid their true value</a>. Even if they were, they would have even worse capital efficiency than sales: they require capital to be locked up by bidders. (Some platforms have even proposed paying bidders <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.coindesk.com/defi-yield-farming-comp-token-explained">DeFi yield</a> on locked up bid capital to mitigate lockups.) In an auction, for every dollar of valuation, the mechanism might require multiple dollars of worth of bidding.</p></li><li><p><strong>The Fractionalization Mechanism.</strong> NFT fractionalization, as seen in the pioneering work of <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://niftex.com/">Niftex</a>, is the first innovation toward creating capital efficiency in an NFT price discovery context. Together with other approaches like <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://ark.gallery/">Ark</a>, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://wrappedpunks.com/">Wrapped Punks</a>, <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.wrappered.com/">WG0</a>, and <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://nftx.org/">NFTX.org</a>, fractionalization shards one or more nonfungibles into an ERC20 currency which can gain liquidity on a DEX or centralized exchange. Anyone can buy any amount of the currency in order to help establish the overall valuation, bringing down the cost of valuation for an individual user (but not necessarily in the mechanism as a whole). With fractionalization also come the <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.niftex.com/the-buyout-clause-in-depth/">challenges of stakeholder governance</a> and managing a proliferation of ERC20 assets corresponding to a large universe of nonfungibles.</p></li></ol><p>This discussion of existing approaches motivates the question of whether there exist approaches that dramatically improve capital efficiency for the price discovery problem. We cover some of these approaches in the next section.</p><h2 id="h-capital-efficiency-is-what-makes-price-discovery-disruptive" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Capital efficiency is what makes price discovery disruptive</h2><p>Based on our discussion so far, a basic framework for evaluating price discovery mechanisms is measuring their <em>capital efficiency</em>. Let us define <em>P(x)</em> as the discovered price of a good <em>x</em> and <em>C(x)</em> as the total cost expenditure from participants required to price it. Then we can broadly define the price discovery efficiency of a mechanism as, informally, <em>E = P/C</em> across a particular set of assets. In this framework, sale mechanisms always have an efficiency of <em>E = 1</em>, auction mechanisms have <em>E ≤ 1</em>, and fractionalization deserves a more nuanced analysis than we should do here, with <em>E ≥ 1.</em> The question is whether we can do much better than fractionalization, and the answer is yes.</p><figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/054817bb81fe209e2eedf54fcd75d8575a1609909d67cd968404d5ec36e2f076.png" alt="NFTBank.ai tracks estimates for virtual real estate, like Decentraland parcels." blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="">NFTBank.ai tracks estimates for virtual real estate, like Decentraland parcels.</figcaption></figure><ol><li><p><strong>Price Computation.</strong> While auctions are great for subjectively-valued goods, collectibles markets are often pricing on scarcity and well-defined properties. Some assets, like CryptoKitties and virtual real estate, might have pricing that can simply be computed. <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://nftbank.ai/">NFTBank.ai</a> is one of the first startups to put forward accurate machine learning models for predicting the prices of collectibles based on past pricing of similar or adjacent collectibles. Virtual real estate pricing might succumb to models which take into account past sales and revenue generation possible in the neighborhood. In this context, capital efficiency goes way up because we can think of the mechanism as having a <em>fixed cost,</em> <em>c</em>, of deploying the pricing algorithms. Thus the cost will be amortized across price discoveries and efficiency <em>E = P/c</em> will go to infinity over time. Still, the jury may still be out on whether machine learning will work well for exuberant and subjectively-valued goods, like <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://www.beeple-crap.com/">Beeple art</a>.</p></li><li><p><strong>Expert Networks.</strong> We don’t necessarily need to build a machine learning model in order to achieve a <em>fixed cost of price discovery</em>. Imagine for each pricing, we pay a fixed amount of money to five experts who give us their educated opinions about a good’s fair market value. Capital efficiency improves as the goods appreciate in value. This approach can use centralized services, or incentivized networks of humans, to create appraisals. One concern with using human experts is scale — will we really have enough experts to process the entire potential volume of goods that the NFT space is likely to produce? As we shall see next, this approach is probably best formalized as on-chain oracle networks, which will naturally incentivize agents to play appraisal games and do so efficiently.</p></li><li><p><strong>Peer Prediction Oracles.</strong> The most exciting recent development in oracles is the implementation of peer prediction as an on-chain mechanism, pioneered by <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://upshot.io/">Upshot</a>. <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://medium.com/upshothq/peer-prediction-101-ba267d8b79ee">Peer prediction</a> is a cooperative game that incentivizes participants to honestly answer queries, without requiring data feeds or other sources of objective ground truth. Upshot proposes to <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://medium.com/upshothq/hot-or-not-for-nfts-d5d221f2a452">apply peer prediction, coupled with an ordering algorithm</a>, to create capital efficient price discovery for NFTs. It’s hard to do justice to peer prediction in this small space, but the mechanism by nature is agnostic between objectively priced collectibles or subjectively priced art — the oracle will report an educated consensus. Most importantly, Upshot makes several interesting improvements to capital efficiencies of today’s appraisal games. First, the mechanism’s valuation costs are amortized across a large number of goods, similar to price computation. Secondly, the protocol’s security margin can be assessed from future revenue: if some appraisers are malicious or bad at their job, then the protocol actively reduces their future cash flows by not selecting them for appraisal tasks. Punishing agents by cutting into their future cash flows instead of making them pay for security up front is a big capital efficiency improvement in cryptoeconomic protocols at large. Upshot’s peer prediction will be the first on-chain mechanism widely applicable to NFT price discovery, and NFT pricing will be the first application of Upshot’s protocol to roll out in 2021.</p></li><li><p><strong>Derivative-Implied Pricing.</strong> NFT loans, demonstrated by companies like <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://nftfi.com/">NFTfi</a>, and NFT indices, as contemplated by <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://nftx.org/">NFTX.org</a>, create another vector for NFT price discovery — the pricing implied by derivatives which have nonfungible assets as their underlying. Custom derivatives, such as the right to purchase an NFT in the future or prediction market shares, can potentially create NFT pricing, while delegating the costs of liquidity aggregation to other platforms or mechanisms. It’s still early in this space, and it will be developing over the coming years as the financialization trend continues.</p></li></ol><figure float="none" data-type="figure" class="img-center" style="max-width: null;"><img src="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/dc77aaa974e1381a5405d7c562a6f061d47e5a0fe9fe792951164e399568c1f9.png" alt="Upshot proposes to use peer prediction coupled with a hot-or-not algorithm to efficiently extrapolate NFT pricing." blurdataurl="data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=" nextheight="600" nextwidth="800" class="image-node embed"><figcaption HTMLAttributes="[object Object]" class="">Upshot proposes to use peer prediction coupled with a hot-or-not algorithm to efficiently extrapolate NFT pricing.</figcaption></figure><h2 id="h-implications-of-nonfungible-liquidity" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Implications of nonfungible liquidity</h2><p>Overall, the potential for capital efficient price discovery mechansims has profound implications for the liquidity of existing NFTs and, by extension, any illiquid asset which can be formulated as an on-chain nonfungible. We are likely to see a whole ecosystem of <em>appraisal games</em> develop around the price discovery problem in the years to come. We have also seen that some approaches are more suitable for objectively priced goods, and some more suitable for subjectively priced goods.</p><p>Here are some practical applications of price discovery.</p><ol><li><p>A creator will be able to produce a work and an efficient market of appraisal games will offer liquidity for that work in a completely automated manner. As such, this would be an utterly disruptive mechanism for monetizing creativity, content, and digital goods.</p></li><li><p>Oracle-based pricing will be used to valuate portfolios and collections of nonfungibles to discover new value.</p></li><li><p>“Instant pricing” of NFTs can be used to create a lower bound at which holders can always liquidate their assets. <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://neolastics.com/">Neolastics</a> have proposed mechanisms along similar lines, and other price discovery might be applicable across a wider range of goods.</p></li><li><p>Any application which uses NFTs as collateral can rely on on-chain pricing to control risk. For example, a lending protocol can set liquidation margins based on automated pricing. Or, in a more technical application, optimistic rollups can potentially use this mechanism to lower “roll-off” costs for NFTs from layer 2.</p></li><li><p>Investors can make purchase decisions faster and more efficiently with suggested prices.</p></li><li><p>We can create decentralized NFT indices backed by the security of appraisal games instead of trust or collateral. This can create major efficiencies for investors seeking exposure to the NFT space but hesitant to evaluate on an asset by asset basis.</p></li></ol><p>I would love to hear your thoughts and feedback on this fascinating and developing space. Follow me on Twitter <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://twitter.com/jbrukh">@jbrukh</a>.</p><h2 id="h-references" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">References</h2><ol><li><p><em>[Emmons]</em> <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://medium.com/upshothq/an-efficient-price-mechanism-for-nfts-efb078dde580"><em>An efficient price mechanism for NFTs</em></a></p></li><li><p><em>[Brukhman] Twitter thread on #</em> <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://twitter.com/jbrukh/status/1341962949976535040"><em>NFTLiquidity</em></a></p></li><li><p><em>[Ausubel, Milgrom]</em> <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://web.stanford.edu/~milgrom/publishedarticles/Lovely%20but%20Lonely%20Vickrey%20Auction-072404a.pdf"><em>The lovely but lonely Vickrey auction</em></a></p></li><li><p><em>[Brukhman]</em> <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.coinfund.io/all-digital-content-is-going-on-chain-ae26a7071657"><em>All digital content is going on-chain</em></a></p></li><li><p><em>[Berenzon]</em> <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://medium.com/bollinger-investment-group/constant-function-market-makers-defis-zero-to-one-innovation-968f77022159"><em>Constant function market makers: a zero to one innovation</em></a></p></li><li><p><em>[Hubert]</em> <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://blog.niftex.com/the-buyout-clause-in-depth/"><em>Niftex: The buyout clause in depth</em></a></p></li><li><p><em>[Emmons]</em> <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://medium.com/upshothq/peer-prediction-101-ba267d8b79ee"><em>Peer prediction 101</em></a></p></li></ol><p><strong><em>Disclaimer: The content provided on this site is for informational and discussion purposes only and should not be relied upon in connection with a particular investment decision or be construed as an offer, recommendation or solicitation regarding any investment. The author is not endorsing any company, project, or token discussed in this article. All information is presented here “as is,” without warranty of any kind, whether express or implied, and any forward-looking statements may turn out to be wrong. CoinFund Management LLC and its affiliates may have long or short positions in the tokens or projects discussed in this article.</em></strong></p>]]></content:encoded>
            <author>brukhman@newsletter.paragraph.com (Jake Brukhman)</author>
        </item>
    </channel>
</rss>