• AI Personas built from deep interviews and social data

• AI‑powered interviews with those personas

• Long‑form reasoning that takes minutes, not milliseconds

• A Fast Insight mode that turns research into podcast‑ready narratives

• Logs and “Nerd Stats” that expose how each study was run​

Why not just build another analytics dashboard?
Why tie together NFTs, Web3 communities, consumer psychology, and AI research?

This article explains the guiding principles behind atypica.ai’s design—and why a crypto‑native, research‑driven team ended up building a platform that looks more like a subjective world simulator than a traditional market research tool.

Principle 1: Business Problems Are Subjective, Not Just Statistical

Our starting point is simple:

Business and social problems are “wicked problems”—
they don’t have clean, purely rational solutions.​

People make decisions based on:

• Stories they tell themselves

• Emotions they can’t fully articulate

• Cognitive biases and social context

Traditional analytics model the objective world: events, transactions, clickstreams.
They’re essential—but they miss the inner layer: how consumers perceive the world.

Atypica.ai is designed to model this subjective world:

• AI Personas capture how certain types of people think and feel.

• AI Interview tools let you talk to those personas as if they were research participants.

• Long‑form reasoning connects the dots between what people say, what they feel, and what they do.​

We didn’t set out to build yet another dashboard.
We set out to build a system that can hold and reason about human interiority.

Principle 2: Think in Agents, Not Averages

In many analytics systems, everything collapses into “the average user.”
In our experience, there is no such person.

Earlier work in Web3 and NFT communities made this obvious:

• One wallet might be a long‑term, values‑driven collector.

• Another is a fast‑moving speculator.

• Another is a builder who cares more about community ethos than price.​

The same is true in every market:

• Some consumers buy for status, others for safety, others for experimentation.

• Some care deeply about privacy; others trade it for convenience without a second thought.

So atypica.ai is built on consumer agents, not aggregates:

• Each AI Persona has a distinct worldview and decision style.

• Personas can be constructed from deep interviews (Human AI Personas), social data, or your own transcripts.​

• Studies involve multiple personas, so you see contrasts and conflicts, not just “one truth.”

This agent‑based approach lets you ask:

• “What would this type of person do?”

• “How would different personas react to the same product or message?”

• “What trade‑offs are acceptable for one segment but unacceptable for another?”

It’s closer to how real strategists and researchers think—just made systematic and scalable.

Principle 3: Transparency Is a Feature, Not an Afterthought

Coming from a world where smart contracts, MerkleTrees, and on‑chain transparency matter, we’re allergic to black‑box “AI magic.”​

Blockchains taught us:

• If something is important, make it verifiable.

• Don’t just claim fairness or rigor—show the evidence.

Atypica.ai inherits this ethos:

• Each study has a clearly defined pipeline: clarify problem → design tasks → gather social signals → build personas → interview → summarize → (optionally) generate a podcast.​

• The platform logs:

  • Which tools were used

  • Which personas were involved

  • How many tokens were consumed

  • How long the reasoning took​

• “Nerd Stats” act like a proof of work for the AI side of the research.​

We don’t claim that transparency solves everything—but it makes it easier to:

• Compare studies

• Audit questionable outputs

• Build trust with teams that rely on the results

In a space where “just trust the AI” is still a common message, we prefer: “Here’s how it actually worked.”

Principle 4: Narratives Are Deliverables, Not By‑Products

We’ve all seen research die in slide decks.

One of the lessons from both Web3 community building and traditional research is:

If you don’t tell a good story about your findings,
the insights won’t change behavior.

That’s why atypica.ai includes Fast Insight:

• A mode that starts from a research brief

• Runs a constrained but deep study

• And outputs a podcast‑ready narrative: script + audio​

We treat narrative as a first‑class output because:

• Stories travel inside organizations better than dashboards.

• People remember a well‑told episode more than they remember charts.

• Opinionated analysis (“Here’s what matters most and why”) is what actually unlocks decisions.

Fast Insight isn’t just a text‑to‑speech feature.
It’s an entire pipeline engineered for audio‑first storytelling, with planning, deep research, and synthesis tuned to that format.​

In other research modes, we still think narratively:

• Reports explain “how we got here” and “what this implies,” not just “what the numbers say.”​

Principle 5: Research Should Be Continuous, Not One‑Off

Traditional research projects are often treated as events:

• Commission study → wait weeks → present → move on.

But humans, markets, and culture are moving targets.
Web3 amplified this feeling: markets shift in days, narratives in hours.

Atypica.ai is built for continuous research:

• Persona ecosystems grow and update over time as new interviews and data come in.​

• Research sessions are tracked with Analyst objects that preserve context, logs, and summaries.​

• You can:

  • Revisit previous studies

  • Ask follow‑up questions

  • Re‑interview the same personas under new scenarios

In practice, this means:

• A launch research study from last year doesn’t vanish—it becomes a starting point for this year’s iteration.

• A persona you “met” in one project can appear again in another, allowing continuity of understanding.

The goal isn’t to replace all traditional research, but to create a living system where insight compounds rather than resetting to zero every quarter.

Principle 6: Use the Frontier to Understand the Mainstream

Our background includes time spent exploring the edges:

• Crypto and NFT communities

• Experimental social and cultural projects

• Topics like privacy‑first products, decentralized coordination, and new forms of digital ownership​

These are not niche curiosities; they’re stress tests for consumer understanding.

If your research tools can make sense of:

• Why people buy JPEGs in bear markets

• Why some communities stay coherent while others implode

• Why narratives about “freedom,” “fairness,” or “clean money” gain traction​

…then those same tools are powerful on more “normal” topics:

• Switching banks or phones

• Choosing between luxury brands

• Deciding whether to trust AI in everyday products​

So we intentionally use atypica.ai on frontier topics—not just because we care about them, but because they sharpen the system’s ability to model complex, value‑laden decisions that show up everywhere.

What This Means If You’re Using atypica.ai

You don’t have to care about NFTs, MerkleTrees, or crypto politics to get value from atypica.ai.

What you get, in practical terms, is:

• A research platform that:

  • Treats your questions as starting points for real reasoning

  • Talks to AI Personas that behave like consistent consumers

  • Produces insights that are transparent, narratable, and agent‑aware​

• A way to:

  • Explore “what if” scenarios before committing resources

  • See how different types of people might react to the same product or message

  • Turn complex topics into stories your team can actually use​

Under the hood, yes, there’s a lineage that runs through Web3 experiments, deep interviews, and agent‑based thinking.
But you don’t need to follow that history to benefit from the outcome.

You just need to care about a simple thing:

Understanding why people decide the way they do—
deeply enough that your products, campaigns, and strategies actually resonate.

That’s what atypica.ai is built to help with.

They’re easy to make, easy to present—and almost impossible to use in real decisions.

Atypica.ai takes a radically different approach.
Instead of static profiles, it builds Human AI Personas: simulated consumers constructed from real data and deep interviews, who can hold conversations, reveal their thinking, and stay consistent across studies.​

This article explains how those personas are created, why they behave more like “living agents” than slides, and how they power a new kind of consumer research.

The Problem with Traditional Personas

Classic personas evolved as a storytelling tool for internal alignment.
But in practice, they suffer from three chronic issues:

1. They’re not grounded
   Many personas are based on incomplete or biased qualitative input, then over‑generalized into archetypes.

2. They’re static
   Once created, they rarely get updated. The world moves; personas stay frozen.

3. They can’t answer new questions
   A slide can’t react when you ask, “How would you feel about this new pricing model?” or “What if we changed the onboarding flow this way?”

Atypica.ai’s Human AI Personas are designed to fix this by treating personas as model‑backed agents, not documents.

Building Personas from Deep Interviews

One of the core methods behind Human AI Personas is deep interviewing:

• Each interview lasts about 1–2 hours.

• AI asks follow‑up questions, adapting to the participant’s answers.

• The process generates on the order of 5,000–20,000 words—
  essentially a short biography of that person’s worldview.​

During these interviews, atypica.ai captures:

• Personal background and life context

• Values and core beliefs

• Decision patterns (how they weigh trade‑offs)

• Emotional triggers and sensitivities

• Specific experiences with products, brands, or categories

This material is then used to construct a high‑fidelity persona model:

• The persona “remembers” important parts of its story.

• It exhibits stable attitudes and preferences.

• It shows recognizable patterns of reasoning across topics.

In other words, if you ask the same persona about different products or scenarios, you get answers that “feel like” they’re coming from the same human mind.

Scaling Up: From Individuals to an Agent Ecosystem

Atypica.ai doesn’t stop at a handful of personas.
Over time, the team has built a large ecosystem:

• Around 300,000 synthetic consumer agents built from social data, covering diverse demographic and behavioral patterns.​

• Around 10,000 high‑precision agents derived from in‑depth interviews and professional cognitive modeling.​

• Private persona libraries that customers can create from their own interview projects, never shared outside their organization.​

What emerges is a living panel:

• Some personas are broad, category‑defining types.

• Others are extremely specific: “young parents in Tier‑1 Chinese cities with high digital affinity,” “Gen Z luxury shoppers with strong sustainability values,” “privacy‑sensitive professionals in Europe,” etc.​

Unlike traditional panels:

• You don’t need to recruit them anew for each project.

• They don’t get tired or bored.

• They can be invited, combined, and reused across many studies.

Using AI Personas as Research Participants

When you run a study on atypica.ai, the platform doesn’t just produce answers from nowhere. It calls relevant personas into the conversation.​

A typical flow might look like:

1. You pose a research question
   e.g., “How would different types of Gen Z professionals perceive a new AI productivity tool that records meetings and summarizes them?”

2. Atypica.ai selects personas

   • A persona representing early adopter knowledge workers

   • A persona representing privacy‑conscious corporate employees

   • A persona representing skeptical creatives who resist “productivity tools”

3. AI Interview sessions begin
   The system runs structured interviews with each persona:

   • Asking open questions

   • Probing for concerns and excitement

   • Testing reactions to specific features and messages​

4. Insights are synthesized
   The long‑form reasoning engine then compares and contrasts persona responses:

   • Which segments are excited vs. anxious?

   • What trade‑offs are they willing to accept?

   • How do they interpret the same value proposition differently?​

You end up with something close to what multiple focus groups might reveal—but generated in minutes and deeply tied to consistent persona logic.

Multiverse Thinking: “What If” Simulations at Scale

Because personas are reusable, you can treat them like nodes in a multiverse experiment:

• Ask: “What if we positioned our product as security‑first instead of convenience‑first?”

• Or: “What if we removed this feature that power users love but casual users ignore?”

Then:

• Run the same set of personas through multiple scenarios.

• Compare their reactions, not just within a single moment, but across different futures.​

This is particularly valuable in:

• Category creation
  When you’re not sure yet which mental model will resonate.

• High‑risk changes
  When altering pricing, data practices, or core UX might alienate some segments.

• Emerging culture spaces
  Such as Web3, creator economies, or new types of hybrid online/offline experiences.

Instead of asking, “What’s the average preference?”, you ask:

“In the multiverse of plausible consumers, which types thrive under which decisions?”

Human AI Personas vs. Synthetic Only Personas

It’s tempting to build personas purely from synthetic modeling (e.g., clustering social data).
Atypica.ai does use synthetic personas, but it places special emphasis on the Human AI Persona layer:

• Personas whose foundation is deep interviews with real people.​

• Cognitive modeling done by researchers, not just unsupervised algorithms.

• Continuous validation by comparing persona behavior against new interview data and social signals.

This human‑grounded layer helps avoid some common pitfalls of purely synthetic agents:

• Over‑smoothed attitudes that feel generic.

• Incoherent value systems that don’t match real lived experience.

• Overfitting to surface sentiment instead of deeper logic.

By mixing human‑grounded personas with synthetic breadth, atypica.ai builds an agent ecosystem that is both rich and anchored.

How Teams Use Human AI Personas in Practice

A few concrete patterns:

1. Message testing across subtle segments

   • You can test variants of copy, visuals, and framing against multiple personas.

   • Instead of “one size fits all,” you see which angles resonate with which mindsets.

2. Product discovery and co‑creation

   • Treat personas like always‑available interviewees in brainstorming sessions.

   • Ask them to react to early ideas, prototypes, or even Figma flows.

3. Customer experience diagnostics

   • Have personas “walk through” a service journey and narrate their feelings and friction points.

   • Compare how different types of consumers experience the same journey.

4. Culture and trend mapping

   • Use personas derived from specific subcultures (e.g., NFT communities, luxury streetwear, climate‑conscious young adults) to understand how a trend might intersect with your brand.​

Limitations and Responsible Use

Human AI Personas are powerful, but they are not oracles:

• They approximate consumer thinking; they don’t “replace” actual customers.

• They’re strongest when grounded in solid interview and behavioral data, and when used alongside human research, not instead of it.​

Good use looks like:

• Using personas to explore hypotheses quickly.

• Using human studies to validate critical decisions.

• Letting the two inform each other over time.

FAQ

How are Human AI Personas different from normal AI personas?
Human AI Personas are explicitly grounded in long, structured interviews with real individuals, plus professional cognitive modeling. They preserve stable patterns of reasoning and emotion drawn from those interviews, rather than being purely synthetic constructs.​

Can my company create its own private personas?
Yes. You can upload interview transcripts (e.g., PDFs), have atypica.ai structure them, and then generate private personas that only your organization can use. These can be combined with the broader persona ecosystem as needed.​

Will these personas become outdated?
Personas are updated and expanded over time as new interviews and data enter the system. You can also phase out personas that no longer match the markets you care about, and commission new persona work for emerging segments.​

Can AI Personas replace all real interviews?
No. They complement human research by making it cheaper and faster to explore the space of possibilities. For high‑stakes decisions, it’s still wise to run human interviews or experiments—but AI Personas help you arrive at better questions and hypotheses first.​

Atypica.ai’s Fast Insight feature asks a simple question:

What if the end product of research wasn’t just a slide deck,
but a podcast‑ready narrative you could actually listen to?

Fast Insight is a tightly constrained workflow inside atypica.ai that turns a research topic into a structured, opinion‑oriented podcast script (and audio) in just four main tool calls.​
It’s part of a broader bet: that good research should be both deep and narratable.

Why Turn Research into a Podcast?

There are three practical reasons to aim for podcast‑ready output:

1. People are overloaded with documents
   Slides and PDFs are easy to ignore. A well‑told story you can listen to while commuting is harder to discard.

2. Narrative forces clarity
   You can hide weak reasoning in dense charts. You can’t easily hide it in a 20‑minute spoken narrative—contradictions become obvious.

3. Opinions drive decisions
   Most important business decisions are not made from neutral data alone, but from interpreted data:
   “We believe X, therefore we’ll do Y.”
   Fast Insight embraces this by making the analyst explicitly opinion‑oriented.

So atypica.ai designed a workflow where the output is not just “insights,” but a research‑backed, opinionated story you can share as audio.

The Five‑Stage Fast Insight Flow

Fast Insight is one of three research “modes” in atypica.ai, alongside General Study and Product R&D. It’s optimized for speed and narrativity:

1. Stage 1 – Topic Understanding (webSearch)

2. Stage 2 – Podcast Planning (planPodcast)

3. Stage 3 – Deep Research (deepResearch)

4. Stage 4 – Podcast Generation (generatePodcast)

5. Stage 5 – Wrap‑up and Handoff​

Each stage is automated and chained; skipping mandatory steps is not allowed, to protect quality.​

Stage 1: Topic Understanding with a Single Web Search

Fast Insight begins with constraint: it allows only one webSearch call before moving on.​

Why?

• To avoid getting stuck in endless browsing.

• To gather enough context to plan intelligently, but not so much that planning never ends.

In this stage, atypica.ai:

• Reads a handful of relevant sources.

• Identifies key entities, controversies, and recent developments.

• Builds a rough mental map of the topic: who’s involved, what’s at stake, what’s changing.​

This becomes the “raw context” for the next step.

Stage 2: Podcast Planning with planPodcast

Next, atypica.ai calls a specialized tool, planPodcast, powered by a planning‑oriented model (Gemini 2.5 Pro).​

The goal here is not to research, but to design a content strategy:

• Pick the most compelling angle for listeners.

• Break the episode into segments (introduction, context, main arguments, counterpoints, implications, closing).

• Decide where to insert:

  • Key data points

  • Quotes or examples

  • “Opinion beats” where the analyst takes a stand​

Fast Insight’s analyst configuration is explicitly set to opinionOriented, meaning:

• The system is allowed—and expected—to say, “Here’s what I think is happening and why,”
  rather than only describing facts.​

This planning step also establishes:

• topic: a refined version of the user’s brief

• How the upcoming deepResearch stage should focus its energy (which sub‑questions matter most)​

Stage 3: Deep Research with deepResearch

Once the plan is set, atypica.ai invokes deepResearch, a multi‑step, long‑running process dedicated to actually understanding the topic.​

Under the hood, deepResearch:

• Uses advanced AI models to combine:

  • Web search

  • X (Twitter) search

  • Prior studies or context where available

• Collects:

  • Key arguments and counter‑arguments

  • Data points and trends

  • Representative quotes from multiple perspectives​

This is where Fast Insight does real work:

• It doesn’t just scrape a few headlines.

• It builds a studySummary: a structured distillation of all the research, stored on the Analyst object.​

Because deepResearch is allowed to take minutes rather than milliseconds, it can:

• Follow chains of reasoning

• Cross‑check claims

• Resolve contradictions where possible

By the end of this stage, atypica.ai has the raw material needed for a thoughtful podcast: facts, patterns, tensions, and emerging viewpoints.

Stage 4: Podcast Generation with generatePodcast

With studySummary and the planned topic in hand, Fast Insight calls generatePodcast.​

This tool:

• Loads the deep research summary.

• Aligns it with the planned structure and tone.

• Writes a full podcast script, including:

  • Host intro and framing

  • Segment-by-segment exposition

  • Opinionated commentary (“Here’s what likely matters most”, “Here’s where I disagree with the mainstream narrative”)

  • Closing thoughts and open questions

It then optionally generates audio and returns a podcastToken, which lets you access:

• The script

• The audio file (via an audioObjectUrl in many cases)​

The result is a piece of content you could:

• Publish as an internal research briefing

• Share with your team or clients

• Adapt into a written article or newsletter issue

Stage 5: Research Closure and Handoff

The final stage is intentionally minimal:

• The system tells you the research is complete.

• It hands you the podcastToken and points you to where you can listen.​

• It avoids dumping the entire research conclusion into the chat again.

Why avoid repeating the research details?

• Because the podcast itself is the primary artifact.

• Because Fast Insight is designed to ship a clear narrative asset, not re‑explain everything in text.

You can always inspect the logs and Nerd Stats if you want to see the underlying process.​

Constraints by Design: Why Fast Insight Is So Strict

Fast Insight comes with hard constraints:​

• Only one webSearch before planning

• No skipping essential tools (webSearch → planPodcast → deepResearch → generatePodcast)

• No continued research after the podcast is generated

• A default maximum of four major steps (matching the tool chain)

These constraints do three things:

1. Prevent analysis paralysis
   The system is nudged to move from context → plan → research → narrative instead of looping forever in early stages.

2. Protect quality
   Forcing deepResearch ensures the podcast isn’t just based on surface‑level reading.

3. Make cost and time predictable
   With a bounded number of tools and steps, token usage and latency are easier to estimate and track.​

This is a good example of where crypto‑native thinking shows up again:
designing processes that are both expressive and bounded, so they remain robust.

When to Use Fast Insight vs. Other Research Modes

Fast Insight is not for every question. It shines when:

• You need a narrative explanation of a topic for stakeholders.

• You want to quickly explore an angle, trend, or controversy and share it as audio.

• You care about opinionated analysis, not just neutral summaries.

Other times, you might prefer atypica.ai’s:

• General Study mode for comprehensive, multi‑tool research across a wide range of methods.

• Product R&D mode for product‑specific questions and experimentation.​

You can also use Fast Insight as a second pass on an existing study:
run a deeper General Study first, then ask Fast Insight to turn those findings into a podcast for broader consumption.

Why Fast Insight Matters

Fast Insight is more than a convenience feature. It encodes a belief:

• That good research should show its reasoning (through transparent logs),

• be deep enough to stand up to scrutiny (through long‑form research),

• and be story‑shaped so people can actually internalize it (through podcast scripts and audio).​

For teams that constantly struggle to get stakeholders to read research, this changes the dynamic:

• Instead of pushing PDFs, you can send a link to “this 18‑minute episode summarizing what we learned about Gen Z and luxury retail” or “this 22‑minute briefing on privacy‑first fintech in Europe.”

• Instead of asking people to memorize charts, you give them stories they can retell.

FAQ

Is Fast Insight just text‑to‑speech on top of a normal report?
No. The entire pipeline is designed with a podcast as the target format. The planning, deepResearch, and generatePodcast steps work together to produce a script that sounds like a human analyst speaking, not a PowerPoint read aloud.​

Can I customize tone and style?
Today, Fast Insight focuses on an opinionOriented analyst style by default. Future iterations are likely to add more control over tone, pacing, and persona of the “host,” while retaining the structured research pipeline.​

What languages does Fast Insight support?
It supports at least Chinese and English, with careful handling of streaming character output for a smooth experience in both languages.​

How does this relate to the rest of atypica.ai?
Fast Insight sits on top of the same foundation as other atypica.ai research modes: AI Personas, structured tools, and long‑form reasoning. It’s just optimized for a very specific output format: a podcast you can listen to and share.​

What’s usually missing is an answer to a deeper question:

Can AI actually help us understand how people think, feel, and decide—
or is it just another way to summarize dashboards?

Atypica.ai takes a very different stance.
It treats market research not as data aggregation, but as modeling subjective worlds: the internal narratives, emotions, and cognitive biases that drive consumer choices.​

This article explains how atypica.ai does AI market research, from raw social signals to structured insights—and why its roots in crypto‑native thinking make it unusually good at this job.

Why Traditional Market Research Struggles Today

Traditional research workflows were designed for a slower world:

• Long recruitment cycles and rigid screening criteria

• Moderated focus groups and expensive in‑depth interviews

• Weeks or months between “we have a question” and “we have an answer”

That approach faces at least three problems now:

1. Speed mismatch
   Product, marketing, and social environments move much faster than legacy research timelines.

2. Shallow understanding
   Surveys and dashboards often reduce complex motivations to checkboxes and single‑number KPIs.

3. Fragmented data
   Social listening, UX research, CRM data, and brand tracking sit in disconnected tools and teams.

Atypica.ai was built to tackle these issues by combining long‑form AI reasoning, AI Personas, and structured workflows that feel like having an always‑on research team.​

Step 1: Start with a Specific Decision Question

Atypica.ai is not a general chatbot. It works best when you treat it like a senior researcher and give it a clear brief, such as:

• “Why are young professionals switching from traditional banks to app‑based fintechs in Europe?”

• “Which storytelling angles for our new beverage will resonate with Gen Z in Tier‑1 Chinese cities?”

• “What kind of person is likely to adopt a privacy‑first browser, and why?”​

The system’s internal pipeline then kicks in:

1. Clarify the problem
   It unpacks your question into sub‑questions and working hypotheses.

2. Design research tasks
   It decides which tools and personas to use: web search, social signals, existing personas, new persona construction, AI interviews, etc.​

This mirrors what a human research lead would do—just much faster.

Step 2: Mine Social and Behavioral Signals

Instead of limiting itself to a static dataset, atypica.ai uses web search and social signals as live context:

• It performs scoped web searches to capture recent conversations, trends, and emerging clusters of opinion.​

• It can incorporate prior studies or interview transcripts you upload, so your proprietary knowledge becomes part of the reasoning.​

This isn’t just about scraping keywords. The goal is to understand:

• How people frame the problem in their own words

• Which metaphors, anxieties, and hopes keep recurring

• Where opinions cluster and where they diverge

Those patterns become the raw material for building AI Personas and study hypotheses.

Step 3: Build AI Personas That Reflect Real Consumers

Atypica.ai’s most distinctive move is to turn raw signals into AI Personas—consumer agents that behave consistently across questions.​

There are several sources:

• Social‑data Personas
  Built from large‑scale analysis of public conversations (e.g., about a category, brand, or lifestyle), capturing shared preferences and attitudes.

• Interview‑based Personas
  Derived from long interviews, where each session generates 5,000–20,000 words of transcript. These personas feel like “biographies encoded as agents”—they know their own backstory, values, and habits.​

• Private Personas
  Created by uploading your own interview PDFs and letting atypica.ai structure and model them. These become proprietary assets only you can access.​

The result is a library of AI Personas that:

• Think and speak like actual consumer types

• Can be re‑used across studies (“urban Gen Z early adopters”, “risk‑averse parents in Tier‑1 cities”, “Web3‑curious but skeptical professionals”, etc.)

• Provide continuity over time, allowing you to see how the same persona reacts to different ideas or campaigns

Step 4: Run AI‑Powered Interviews at Scale

Instead of just summarizing existing text, atypica.ai talks to its personas using AI Interview tools:

• It asks open‑ended questions.

• It follows up when answers are vague or contradictory.

• It probes for deeper motives:

  • “Why is that important to you?”

  • “What would make you change your mind?”

  • “How would you explain this choice to a friend?”​

Because personas are agents with stable traits:

• A “price‑sensitive pragmatist” persona will consistently push back on premium positioning.

• An “aesthetic‑driven collector” persona will be more sensitive to design and cultural signals.

• A “privacy maximalist” persona will analyze a product’s data practices in detail.

You can run dozens or hundreds of such interviews in minutes, exploring a wide matrix of personas × scenarios without recruitment bottlenecks.

Step 5: Long‑Form Reasoning and Synthesis

Once interviews and data gathering are complete, atypica.ai switches into long‑form reasoning mode:

• It spends 10–20 minutes (or more, depending on settings) “thinking” through the material.​

• It looks for:

  • Repeated patterns across personas

  • Contradictions that need resolving

  • Latent tensions (e.g., desire for convenience vs. fear of surveillance)

  • Under‑served niches and opportunity spaces​

The output is not just a bullet list of “insights,” but a structured narrative that often includes:

• Clear articulation of the core consumer problem as they perceive it

• Key consumer segments and what differentiates them

• Emotional and cognitive drivers behind adoption, churn, or indifference

• Implications for product, messaging, and go‑to‑market

For teams that want a more shareable artifact, this same summary can feed into Fast Insight to generate a podcast episode explaining the research as if you were listening to a thoughtful analyst talk it through.​

Step 6: Transparent Logs and “Nerd Stats”

One of the most underrated features in atypica.ai is how it makes the research process itself visible:

• Each study has a log of:

  • Tools used

  • Steps executed

  • Time spent

  • Tokens consumed

  • AI Personas involved​

• These details are presented as “Nerd Stats,” functioning as a kind of proof of work for the research.

This matters because AI research can otherwise feel like magic. With atypica.ai, you can:

• See how much effort went into a given study.

• Compare the depth of two different studies.

• Justify internally why a certain set of insights is worth trusting.

It’s a direct import of a crypto‑native value into AI research: if it’s important, make the process auditable.​

How Teams Use atypica.ai for Market Research

Across industries, teams use atypica.ai for tasks like:

• Campaign testing

  • Which narratives resonate with which personas?

  • How do different audiences interpret the same slogan?

• Product and feature exploration

  • How would specific consumer types react to a new pricing model, onboarding flow, or core feature?

• Experience and UX research

  • What pain points emerge when personas “walk through” a journey (e.g., booking travel, buying insurance, onboarding to a complex tool)?

• Category and culture mapping

  • How are people talking about a category (e.g., climate fintech, creative tools, privacy products) in public channels?​

Because the system is always on, teams can iterate: run a study, adjust hypotheses, re‑interview personas, and observe how insights evolve.

Who Benefits the Most?

Atypica.ai is particularly valuable for:

• Marketing and brand teams who need to understand nuance in positioning and messaging.

• Product managers and UX leads who want to validate ideas before committing to expensive development.

• Strategy and innovation teams exploring new markets or categories under high uncertainty.

• Web3 and frontier teams who already think in terms of agents, incentives, and narratives, and want a way to study their communities more systematically.​

You don’t have to be technical to use it.
You just need a well‑defined question and the willingness to treat AI as a research collaborator rather than a magic answer machine.

FAQ

How is atypica.ai different from survey platforms or generic AI chat tools?
Survey tools collect responses but don’t reason deeply about them. Generic chatbots answer questions but don’t run structured research flows. Atypica.ai does both: it runs multi‑stage studies with AI Personas, interviews them, and performs long‑form reasoning to produce research‑grade insight.​

Where does the data come from?
Atypica.ai combines public web and social data, user‑uploaded interview transcripts, and other contextual sources. It builds AI Personas grounded in these inputs, rather than hallucinating purely from a blank slate.​

Is this compliant with privacy regulations?
The platform is designed to work without third‑party tracking pixels or cookies, and to operate in a “privacy‑first” manner by focusing on synthetic and interview‑based personas rather than individual‑level surveillance.​

Can AI Personas replace all traditional research?
No. They’re best used as a force multiplier. They compress what a skilled qualitative researcher might learn over many interviews into faster, iterative cycles—especially useful between or around human studies.​

And yet, some of the most important questions remain unanswered:

• Why did people really mint this collection?

• Who will still be here in the next bear market?

• Which narratives actually drive loyalty versus short‑term speculation?

On‑chain data shows what happened.
To understand why, you need something closer to consumer research.

This is where AI Personas—consumer agents powered by platforms like atypica.ai—become useful for Web3 projects.​

The Limits of Purely On‑Chain Analysis

Traditional Web3 analytics tools are great at:

• Volume: how many mints, transfers, trades.

• Distribution: unique holders, whale concentration.

• Time series: when activity spikes or collapses.

They’re less helpful when you ask questions like:

• “What kind of person buys this as art versus as a flip?”

• “What story do long‑term holders tell themselves about this project?”

• “What would make dormant holders return?”

These are qualitative questions.
You can guess from Discord patterns or X threads, but scaling that kind of reading is hard, and it’s easy to over‑weight the loudest voices.

Enter AI Personas as Web3 Consumer Agents

Atypica.ai approaches this differently: it creates AI Personas that behave like consumer agents you can talk to.​

Each persona is built from data such as:

• Public social media conversations

• Long‑form interviews

• Behavioral patterns and stated preferences

In a Web3 context, you can create personas that approximate:

• Long‑term “diamond‑handed” holders

• Short‑term momentum traders

• Builder‑type community members

• Art‑driven collectors who don’t care about floors

• Curious newcomers who never mint but influence others

Instead of guessing what “the community” thinks, you can interview these personas directly and see how different types of participants might respond.

A Case Pattern: Analyzing a Pixel‑Art NFT Project

Consider a pixel‑art NFT project with a strong technical story—fair minting via MerkleTree proofs, pre‑generated art, and a small, research‑minded core team.​

Atypica.ai can help such a project in several ways:

1. Build segment‑specific personas

   • One persona might represent early minters who were drawn by the technical write‑up.

   • Another might represent collectors who arrived later via secondary market discovery.

   • A third might represent people who followed the project closely but never minted.

2. Run simulated interviews
   Using AI Interview capabilities, the platform can ask each persona questions like:

   • “Why did you decide to mint (or not mint)?”

   • “Which aspects—art, story, team, mechanics—mattered most to you?”

   • “What would make you commit more deeply to this community?”

3. Map emotional and narrative drivers
   The system then identifies patterns:

   • Fear of missing out vs. genuine alignment with the ethos

   • Attachment to the art style vs. attachment to the builder story

   • Sensitivity to floor price vs. willingness to hold through volatility​

This is similar to what atypica.ai did in studies like “hippyghosts ft. bmrlab”, where the platform was used to analyze NFT communities at a deeper level than on‑chain data alone can offer.​

From Wallet Addresses to “Why” Stories

A critical shift when using AI Personas is moving from address‑level behavior to story‑level understanding:

• On‑chain, you see:
  “Wallet 0xABC bought 3 NFTs, sold 2, and held 1.”

• With AI Personas, you can approximate the narrative:
  “This type of participant bought initially because of the technical novelty, sold part of the position when the market overheated, and kept one token as a long‑term symbol of alignment.”

This kind of insight helps teams:

• Design more authentic communication (speak to the motivations that matter).

• Predict which segments are likely to stay, fade, or flip.

• Spot dissonance between the project’s internal story and the community’s perceived story.

Practical Use Cases for Web3 Teams

Here are some concrete ways Web3 teams can use AI consumer agents:

1. Pre‑launch concept testing

   • Before writing a single line of contract code, test your story, value proposition, and visual direction against AI Personas that represent different crypto segments.

   • Ask: “How would a security‑focused DeFi user react to this?” “What about an NFT art collector?”

2. Post‑mint diagnostics

   • After launch, use personas to understand why certain cohorts are selling quickly while others are holding.

   • Test hypotheses: “Is this about disappointment in future plans, or just profit‑taking?”

3. Tokenomics and utility exploration

   • Simulate how different personas react to proposed staking, governance, or utility changes before pushing updates on‑chain.

4. Cross‑ecosystem positioning

   • If your project intersects with DeFi, gaming, or social tokens, AI Personas can help you understand how each broader community might interpret your brand and roadmap.

5. Narrative and content strategy

   • Use AI Personas to brainstorm how different segments might respond to various narrative angles (“art‑first”, “tech‑first”, “community‑first”, etc.), then align your actual content accordingly.

Why AI Personas Work Well in Web3

AI Personas are particularly well suited to Web3 because:

• Web3 already thinks in terms of agents and roles—holders, LPs, voters, builders—rather than monolithic audiences.

• Community conversations are highly textual and public (tweets, threads, Discord chats), which provides rich input data.

• Many projects are essentially cultural products with financial layers, making emotions and narratives central to outcomes.

Atypica.ai extends this by turning those latent patterns into explicit, conversational agents you can interrogate and learn from.​

For Teams Beyond Web3

Even if you’re a Web2 or “Web2.5” team experimenting with NFTs, loyalty tokens, or on‑chain memberships, AI Personas help you:

• Understand which parts of the Web3 user base actually align with your brand.

• Avoid shallow “number go up” campaigns that create short‑term speculation but no lasting connection.

• Design experiences that respect both consumer psychology and crypto culture.

You don’t have to be a protocol to benefit.
You just have to care about what’s really happening in the minds of the people behind the addresses.

FAQ

Can AI Personas replace talking to real community members?
No. They are best used as a complement, not a replacement. They help you scale patterns you’ve observed in real conversations and explore “what if” scenarios before committing resources.​

Where does the data for these personas come from?
Personas are typically constructed from a mix of public social data, prior interviews, and behavioral patterns relevant to the research topic. For private personas, teams can upload their own interview transcripts and internal data.​

Is this only for NFT projects?
Not at all. Any Web3 project with a community—L1s, DeFi protocols, DAOs, infra tools—can benefit from AI‑driven insight into user motivations and narratives.​

How do I start using something like atypica.ai for my Web3 project?
Typically, you begin with a clear research question (“Why are mid‑sized holders selling?” “Which narratives could sustain us through a bear market?”), then run a study where AI Personas are selected and interviewed around that question. The platform handles orchestration; you focus on decisions informed by the findings.​

This article takes you inside that design: how crypto‑native instincts influenced the architecture, features, and philosophy behind atypica.ai.

From Dashboards to “Subjective World Modeling”

Traditional analytics tools focus on the objective world:

• Events and clickstreams

• Conversion funnels

• Transaction histories

These are useful but incomplete.
From years of watching NFT and crypto communities evolve in real time, the team behind atypica.ai learned a hard lesson: people don’t behave like rational agents, whether they are minting a collection or choosing a SaaS tool.​

So atypica.ai starts from a different premise:

Business research is really about understanding subjective worlds—
the narratives, emotions, and cognitive biases inside people’s heads.

That leads to design choices like:

• Treating each study as a multi‑step process of reasoning, not a one‑shot answer.​

• Representing consumers as AI Personas with distinct voices and mental models, not just as segments in a pie chart.​

• Building flows that look more like how a human researcher works: clarify → explore → talk to people → synthesize → present.​

Composable Architecture: Many Models, One Research Flow

In Web3, composability is a core principle: small, specialized protocols that can be combined to create powerful systems.
Atypica.ai applies that same idea to AI research:

• Claude 3.7 Sonnet acts as the main reasoning engine, handling deep analysis and synthesis for complex questions.​

• Gemini 2.5 Pro handles planning: structuring research flows, designing podcast outlines, and orchestrating steps.​

• Custom tools handle:

  • Web search and X (Twitter) search

  • Persona construction from interviews and social data

  • AI‑powered interviewing

  • Podcast script and audio generation (Fast Insight)​

Instead of asking one model to do everything, atypica.ai coordinates a small “team” of models and tools—each doing what it does best, all hidden behind a single research interface.

Transparent Pipelines: Bringing “Verify, Don’t Trust” into AI

Crypto‑native builders are allergic to black boxes.
In blockchains, everything important is auditable: contract code, transaction history, governance votes.

Atypica.ai brings that ethos into AI research by treating the research workflow itself as something that should be visible and inspectable:

• Studies are broken into named stages—for example:

  • Clarify Problem

  • Design Tasks

  • Browse Social

  • Build Personas

  • Interview Simulation

  • Summarize & Report​

• Each stage logs:

  • Which tools were called

  • Which AI Personas were involved

  • How many tokens were used

  • How long things took​

• The platform exposes “Nerd Stats” so users can see the cost and structure behind every study, not just the final PDF.​

It’s not on‑chain, but the spirit is familiar: don’t just say “we did research”—show how it was done.

AI Personas as the Core Primitive

Many research tools treat personas as static artifacts: a slide with a fake name, stock photo, and a bulleted list of traits.
Atypica.ai treats personas as live agents.

There are several layers:

• Synthetic Personas from social data
  Built by analyzing public conversations and patterns, these personas represent recognizable types in the market—early adopters, price‑sensitive skeptics, enthusiasts, etc.​

• High‑fidelity Personas from deep interviews
  Through 1–2 hour interviews, each generating 5,000–20,000 words of transcript, atypica.ai can turn a real person’s worldview into a reusable persona that behaves consistently across questions.​

• Private Personas from your own data
  Teams can upload interview documents, have atypica.ai structure them, and then create proprietary personas that no one else has access to.​

When you ask a research question, atypica.ai:

1. Identifies which personas are relevant to the topic.

2. Runs simulated interviews with them using AI Interview tools.​

3. Aggregates patterns into a structured answer: a report, a slide‑friendly summary, or even a podcast script.​

This is an agent‑centric view of consumers—the same mental model Web3 builders applied to different wallet types and protocol participants.

Fast Insight: From Research Brief to Podcast in Minutes

One of the most striking features in atypica.ai is Fast Insight, a pipeline that turns a research topic into a podcast‑ready asset in just four main tool calls:​

1. webSearch
   A quick, strictly limited pass over public web content to gather immediate context.

2. planPodcast
   A planning step, powered by Gemini, that selects the most engaging angles, outlines segments, and defines the tone.

3. deepResearch
   A multi‑step, intensive research process that uses advanced models, web search, and X search to collect data, opinions, and trends.​

4. generatePodcast
   A synthesis step that turns the study summary into a full script and audio, returning a podcastToken that links to the content.​

This flow is fully automated, with smooth streaming output and low latency for Chinese and English.
It’s research designed from the outset to end in narrative form, not just a static report.

Resilience: Handling Long Studies Like Long‑Running Protocols

Crypto builders are used to long‑running systems: contracts that live for years, DAOs that keep evolving, and protocols that survive upgrades and forks.

Atypica.ai was designed with similar resilience traits:

• Research sessions are tracked via an Analyst object that stores:

  • Locale

  • The initial brief

  • The emerging topic

  • Study summary

  • Study log

  • Attachments​

• Long‑running studies can continue in the background via backgroundTokens, surviving temporary interruptions.​

• Usage is stored in ChatStatistics, giving a detailed record of time, tokens, and steps—useful both for billing and for understanding how the system works.​

The result: studies feel more like persistent research projects than single chat sessions.

What This Means If You’re Not in Crypto

You don’t need a Web3 background to benefit from this design.
What you get as a user is:

• A platform that takes your questions seriously and allocates real reasoning time to them.​

• Answers grounded in simulated conversations with AI Personas, not just pattern‑matching against a generic corpus.​

• The ability to see how each study was conducted: which tools, which models, which steps.​

Crypto‑native thinking is just the origin story.
The end result is an AI research platform built for anyone who needs to make decisions in complex, human‑driven systems: marketers, product managers, strategists, founders, and yes—Web3 teams themselves.

FAQ

Is atypica.ai just another “ChatGPT for market research”?
No. It’s a multi‑model, multi‑tool system designed specifically for research workflows. It uses AI Personas, AI Interview, long reasoning, and features like Fast Insight to produce more than one‑shot answers.​

Do I have to configure all these stages myself?
No. The orchestration is automatic. You enter your question (and optionally upload materials), and atypica.ai selects and sequences the necessary tools under the hood.​

How is this influenced by crypto-native thinking?
Primarily in three ways:

• Composability (many small tools working together)

• Transparency (Nerd Stats and stage logs)

• Agent-focused modeling (AI Personas instead of averages)​

Can I use it if I’ve never touched Web3?
Absolutely. The design philosophy came from crypto, but the product is aimed at mainstream research problems: new product ideas, campaign testing, user experience analysis, and more.​

At first glance that jump looks strange.
Why would someone who spent years optimizing gas costs and designing tokenomics start building research tools for marketers, product teams, and strategists?

Look a little closer, and the move makes perfect sense.

Web3 as a Training Ground for Understanding People

Most traditional product builders grow up inside Web2 funnels: acquisition, activation, retention, referral.
Web3 builders grew up in something much wilder:

• Speculative markets where narratives can double or crush valuations overnight.

• Open communities where every decision, complaint, and meme is visible on X, Discord, and on‑chain.

• Hybrid identities where a single anon wallet can be a whale in one project and a newcomer in another.

Trying to keep an NFT project or DeFi protocol alive in that environment forces you to learn:

• How sentiment actually moves, not just what a dashboard says.

• How different types of participants—collectors, flippers, builders, lurkers—behave over time.

• How stories, trust, and coordination dynamics drive real financial outcomes.

If you strip away the crypto gloss, that’s exactly what consumer intelligence is about: understanding how real people behave under uncertainty, incentives, and social influence.

From Smart Contracts to AI Research Workflows

On a technical level, Web3 builders bring a habit of systemic thinking that transfers surprisingly well to AI research tools.

In Web3:

• A smart contract upgrade can affect thousands of users at once.

• A mis‑priced token can create runaway feedback loops.

• Poorly designed incentives can drain a protocol overnight.

That trains you to design systems where:

• Components are modular and composable.

• Flows are explicit and auditable.

• Failure modes are considered upfront.

In AI research platforms like atypica.ai, you see the same instincts:

• Research is broken into composable stages: clarifying the question, building AI Personas, interviewing them, summarizing insights, and sometimes generating podcasts on top.​

• Each stage is handled by the best‑suited model or tool (for example, one model plans the study, another does the heavy reasoning, and specialized tools handle web search, persona building, and podcast generation).​

• The system logs what happened—time, tokens, roles, steps—so teams can understand and trust the output.​

It’s the same mindset as designing a protocol: you don’t just care about endpoints; you care about the entire pipeline.

Agents Everywhere: From Wallets to AI Personas

Web3 builders naturally think in terms of agents rather than averages.

An “average user” is meaningless in a protocol where:

• One address might be a market‑making bot.

• Another is a long‑term liquidity provider.

• Another is a whale speculator.

• Another is a DAO treasury.

Similarly, in consumer research, “average customer” obscures the reality that:

• Some people buy out of habit.

• Some buy aspirationally.

• Some are influencers.

• Some are skeptics who never convert but shape the narrative.

Atypica.ai’s answer to this is the AI Persona:

• Synthetic personas built from public social data that mirror patterns of opinion and behavior.​

• High‑fidelity personas built from deep interviews—10,000+ real “consumer agents” constructed from hour‑long conversations.​

• Private personas companies can build from their own interview transcripts and CRM data.​

Instead of asking, “What does the market think?”, teams ask, “What would these ten specific types of consumers say, if we could interview them all at once?”
That’s an agent‑based way of thinking—very natural to someone who spent years reasoning about different wallet types in Web3.

Transparency as a First‑Class Feature

One of the founding values in Web3 is “don’t trust, verify.”
That’s why builders obsess over:

• Contract addresses and verified source code.

• Merkle proofs for airdrops and fair mints.

• Immutable logs of transactions and governance votes.

When those same people build AI tools, they don’t suddenly become comfortable with opaque black boxes.
Instead, they push for:

• Explainable research flows: which tools were used, in what order, with what intermediate results.

• Visible costs: how many tokens were consumed, how long reasoning took, what models were involved.

• Replayable studies: the ability to revisit, tweak, and re‑run a research flow with slightly different prompts or personas.

Atypica.ai’s stage‑by‑stage logs and “Nerd Stats” are concrete examples of this transparency obsession living inside an AI product.​

Why This Matters for Non‑Crypto Teams

If you’re a marketing lead, product manager, or strategist with no interest in Web3, you might wonder:

“Why should I care that my research platform was designed by people from the crypto world?”

You don’t need to care about the crypto part.
You care about the resulting product qualities:

• A tool that takes subjective behavior seriously instead of pretending consumers are rational calculators.​

• An architecture that treats personas as agents, not static slides.​

• A research flow that is fast enough for modern product cycles but transparent enough to trust.​

Crypto‑native teams have simply spent years in an environment where:

• Behavior is public.

• Stakes are high.

• And trust is fragile.

That experience tends to produce AI research tools that are more honest about how messy human decision‑making actually is.

The Hidden Upside: Researching Web3 Itself

There’s also a nice symmetry here.

Platforms like atypica.ai can be used not just by traditional brands but by Web3 projects themselves:

• NFT teams can build AI Personas representing different holder types and ask them, “What would make you stay through a bear market?”

• DAO contributors can explore how different segments perceive governance proposals.

• Protocol teams can test messaging, onboarding flows, and token design with simulated users before shipping.

In some cases, the same builders who learned from their Web3 communities now use AI‑driven research platforms to study those communities back—closing a loop between on‑chain behavior and modeled subjective worlds.​

FAQ

Do you need to understand crypto to use AI tools built by Web3 teams?
No. The crypto background is part of the builder’s story, not a requirement for users. Platforms like atypica.ai are designed for researchers, marketers, and PMs in any industry.​

How is this different from a traditional research agency?
Traditional agencies are human‑only, slow, and often opaque. Crypto‑native AI platforms automate large parts of the process, simulate consumers with personas, and expose detailed logs of how each study was run.​

Is this just hype, or does it meaningfully change research quality?
The key difference is the treatment of subjective factors—emotions, narratives, and social context—combined with agent‑based thinking. That tends to produce insights closer to how people actually behave in volatile, status‑driven markets.​

Can these tools help Web3 projects too?
Yes. Web3 teams can use AI Personas to understand different kinds of holders, test new narratives, and simulate reactions to product or tokenomics changes before pushing anything on‑chain.​

Out of that environment, a new type of research team emerged: crypto-native, research-obsessed, and deeply familiar with what actually moves people when money, memes, and meaning collide.
BMRLab is one of those teams—and atypica.ai is the AI consumer research platform that grew out of their journey.​

From Pixel Ghosts to Real Consumers

Before atypica.ai existed as an AI research product, its builders were involved in Web3 experiments: NFT collections, on-chain communities, and protocol-adjacent projects.

One of the emblematic projects in this lineage is a pixel-art NFT collection that took fairness and technical integrity unusually seriously. The team:

• Pre-generated all 9,999 images.

• Hashed them into a MerkleTree so that the full set was fixed before mint.

• Used the Merkle root in the contract to guarantee that neither rarity nor specific images could be manipulated after seeing who minted what.​

It wasn’t just a generative art drop; it was a statement about transparency.
If you say your mint is fair, you should be able to prove it cryptographically.

That mindset—“don’t just claim, prove it”—is exactly what later informed how this same team approached AI-driven consumer research.

What Web3 Taught Them About People

Running an NFT community at scale forces you to develop instincts that look a lot like advanced consumer research:

• You learn that holders are not one persona.
  There are long-term collectors, short-term flippers, culture-driven fans, and silent whales.

• You learn how narratives beat spreadsheets.
  Floor prices move not only with “fundamentals” but with stories: who joined, what was said, which thread went viral.

• You learn that behavior is public but motivation is hidden.
  On-chain data shows what wallets did—not why they did it.

To keep a community healthy, you need more than dashboards. You need a feel for subjective worlds: fear, hope, belonging, status, boredom, curiosity.
That’s exactly the space where large language models later became useful—not as calculators, but as simulators of human reasoning and conversation.

From On-chain Provenance to Research Provenance

When this crypto-native team decided to build an AI research product, they didn’t start from “how do we make a prettier dashboard?”
They started from:

“If blockchains gave us verifiable transactions, what would verifiable research look like?”

On the NFT side, MerkleTrees and immutable contracts guaranteed that:

• The artwork set was predetermined.

• Anyone could verify that a revealed image matched its committed hash.​

On the AI research side, atypica.ai applies the same spirit:

• Every study records how the AI worked: tools called, time spent, tokens consumed, roles involved.​

• Long-form reasoning is not just a hidden black box; it’s reflected in structured logs—what atypica calls “Nerd Stats.”​

• Instead of claiming “trust us, we used AI,” the platform shows the process behind each report.

Crypto taught BMRLab that process transparency isn’t a nice-to-have—it’s how you earn trust when systems are complex and stakes are real.

Modeling the “Subjective World” Instead of Just Metrics

Traditional analytics focus on the objective side of behavior: events, funnels, click-through rates, conversion.
But the Web3 years showed BMRLab that people don’t act like rational agents, whether they are minting an NFT or choosing a new banking app.​

That led to atypica.ai’s core thesis:

• If physics models the objective world,

• then language models can help us model the subjective world of consumers:
  emotions, narratives, cognitive biases, social context.​

Instead of only aggregating data, atypica.ai:

• Builds AI Personas from social media signals and deep interviews—simulated consumers that preserve individual cognitive patterns.​

• Runs AI-powered interviews with those personas, asking follow-up questions until motivations and mental models become clear.​

• Uses long reasoning chains (10–20 minutes) to generate research reports that read more like a thoughtful analyst than a quick summary.​

Where NFT projects forced the team to understand why anonymous wallets behaved the way they did, atypica.ai generalizes that skill to any consumer context: fintech, luxury, gaming, culture, Web3, and beyond.​

The Loop Closes: Researching Web3 Communities with AI

One particularly interesting turn in this story is that atypica.ai doesn’t just inherit ideas from Web3—it is also used to study Web3 itself.

In case studies like “hippyghosts ft. bmrlab”, the platform is applied to analyze NFT communities:​

• AI Personas are built that reflect different segments of holders.

• Simulated interviews probe their reasons for minting, holding, selling, or simply watching.

• The system maps emotional triggers, narratives, and social structures that can’t be seen in price charts alone.​

This creates a satisfying loop:

• Web3 communities shaped how the team thinks about research.

• Now their AI research platform helps others understand Web3 communities at a much deeper level.

From Ghosts to Personas, From Mints to Insights

Seen from the outside, “NFT project” and “AI research platform” look like two different worlds.
From the inside, the continuity is surprisingly strong:

• Same curiosity: Why do people do what they do when incentives, identity, and narrative all interact?

• Same respect for proof: Don’t just say “fair mint” or “rigorous research”—make the process auditable.

• Same focus on agents: Wallets in NFT land, AI Personas in consumer research; both are lenses on human behavior.

BMRLab’s journey from pixel ghosts and MerkleTrees to AI Personas and long-form reasoning isn’t a hard pivot.
It’s the same problem—understanding complex, subjective decisions at scale—approached with new tools.

If You’re Building at the Edge of Web3 and AI

If you’re exploring how to bring Web3-grade transparency and nuance into your consumer research, atypica.ai is one of the most interesting places to start.​

You don’t need to care about NFTs to benefit from it.
You just need to care about understanding why people decide the way they do—and be willing to let a crypto‑native research engine help you see the subjective world more clearly.

FlashLoan 是一种无抵押借贷。任何人都可以随时从资金池中借出大量资产，用它在区块链上做任何交易，最后把连本带利还给资金池。整个过程需要在一个交易里完成。

闪电贷攻击

1. 2020年10月26日总锁仓量超过10亿美元的 DeFi 项目 Harvest Finance 曝出遭到黑客攻击，已造成大约2400万美元的损失。

2. 2020年11月17日，起源协议 Origin Protocol 稳定币OUSD被爆出遭到闪电贷攻击，Origin Protocol共损失225万美元的 DAI 和100万美元的 ETH。

3. 2021年5月12日，DeFi 质押和流动性策略平台 xToken 遭到攻击，xBNTa Bancor 池以及 xSNXa Balancer 池被耗尽，xToken 损失约2500万美元。

4. 2021年 5 月 20 日，BSC 链上的 DeFi 收益聚合器 PancakeBunny 遭到闪电贷攻击。攻击者利用 PancakeSwap 操纵 LP Token（BNB-BUSDT/BNB-BUNNY）的价格，造成损失 4,500 多万美元。

5. 2021 年 Cream Finance 多次遭受闪电贷攻击，2 月份损失 3750 万美元，8 月份损失 1900 万美元，10月份又损失约 1.3 亿美元。

2022 还没结束，闪电贷攻击还在继续 ...

闪电贷可以用来做什么

市场永远存在大量的套利空间和漏洞，通过一次性借出大量的金额，可以

1. 找到套利空间，进行大规模的低买高卖，获得合理收益；比如资产在交易所 A 的价格低于交易所 B 的价格 0.1%，大规模资金很容易套利。

2. 将大量资金注入市场，操纵物价/币价，扰乱市场价格（预言机），利用极端情况下某些协议出现的漏洞，进行攻击。

如何确保还钱？

这个在以太坊上不是一个问题，因为以太坊是一个单实例的计算机，也是一个单实例的数据库，可以把任何操作放在同一个区块里进行原子操作。执行闪电贷的前提就是所有的交易需要在同一个区块里完成，这个由闪电贷提供者通过程序实现。

直觉

区块链给人最直观的感觉是去中心化的网络，特点是运行节点分散，并且数据永久。但是理解闪电贷的过程让人意识到，在 trustless 的场景下很多事情的做法不同。由于 trust 不需要任何成本，trust 本身是没有价值的，那么传统世界里本身基于 trust 发生的事情，在以太坊上面就会不一样。

在以太坊上，默认信任第三方的假设是不存在的，并且可以随时回滚整个交易，所以默认不信任没关系，但自己要管理好自己的状态。在互联网上，如果没有默认信任第三方，就无法完成任何交互。

以太坊运行

合约执行的特点

1. 状态更新需要合约执行，合约执行需要发起交易，矿工进行分布式记账，有人为之买单

2. 交易是一系列操作的集合，交易成功以后，状态才会改变，交易失败不会改变任何状态

3. 合约不会并行执行，交易串行执行，合约执行也是串行的，所有状态变更都有明确的先后顺序

4. 合约无法自动执行，没有 0 成本的定时任务

5. 合约执行成本高，全量更新等洗数据操作要通过优化过的数据结构来替代

以太坊的状态，包括余额，都是通过记账的方式进行维护，一次记账就是一次状态变更记录，如果记账项目被删除，状态的变更也就会回退。举个例子：

初始状态 a = 0, b = 10
交易 1: a = a + 1, b = b - 1 执行成功 ✅
交易 2: a = a + 1, b = b - 1 执行失败 ❌
交易 3: a = a + 1, b = b - 1 执行成功 ✅
三次交易以后, 状态 a 等于 2，状态 b 等于 8

交易 Transaction

交易 = 执行代码并改变状态，交易的形式确保执行权限和执行费用，一个交易包括：

 from: 交易发起人
   to: 交易对端
value: from 发送给 to 的 ETH 数量
 data: 额外信息
  gas: from 支付给矿工的燃气费

一个转账交易的例子

X 给 Y 转 100 ETH，上面 a 和 b 的例子就是一个转账交易的状态变更过程：from 告诉矿工，把自己的余额减少 100 ETH，在把 to 的余额增加 100 ETH。

 from: X
   to: Y
value: 100ETH
 data: 无
  gas: 0.01ETH

执行的规程如下：

Transaction 1
  |- X 的余额 -100ETH
  |- Y 的余额 +100ETH

如果交易 Transaction 1 执行失败，交易内部产生的所有状态变更都不会被记录，所以不会出现 X 的余额减少但是 Y 的余额增加的情况。除非是这么操作：

Transaction 1
  |- X 的余额 -100ETH
Transaction 2
  |- Y 的余额 +100ETH

交易 Transaction 2 会在 Transaction 1 执行以后再执行，不管 Transaction 1 执行的结果是什么，Transaction 2 都会修改 Y 的余额。转账操作一定要把转账双方余额变更的操作放在一个交易里执行。

一个合约执行的交易的例子

X 执行合约 C，在合约里把自己的名字改成 zhangsan

 from: X
   to: C
value: 0
 data: rename("zhangsan")
  gas: 0.03ETH

地址 C 上面部署着一个合约，合约里实现了 rename 方法，接收一个字符串作为参数，交易执行过程：

Transaction 1
  |- name = "zhangsan"

一个小问题，质押收益如何分配

问题描述

• 有一个资金池，会持续产生收入

• 每产生一笔收入，就会按照当前池子里所有人的份额直接分配收益，人越多，每个人分配到的就越少

• 所有人(Stakeholder)可以在池子里质押(Stake)某个特定代币证明自己的份额

• 可以随时追加或者减少自己的份额，但是收益分配在收入产生的那个瞬间直接发生

很多金融工具都需要这样的运行机制，比如比如活期存款利息，比如公司股权。

举个例子

以区块作为时间单位

1. 区块1，A 和 B 分别质押 300 股和 100 股，A 占 75%，B 占 25%

2. 区块2，进来一笔收益 $100，A 和 B 分别得到 $75 和 $25

3. 区块3，B 追加 200 股，最终 A 和 B 各占 50%

4. 区块4，进来一笔收益 $200，A 和 B 分别得到 $100 和 $100

最终，4个区块过后，A 和 B 的收益分别为 $175 和 $125

要注意的是

收益没法是积累到一定程度然后按照那时候的份额再分配，因为过程中所有人的份额都可以随时调整，合理的情况是质押时间越久，分享的收益就越多。比如上面的情况里，A 和 B 在 4 个区块后，最终的份额都是 50%，如果收益积累了 4 个区块再分配，那 A 和 B 就会各得 $150，显然 A 投资时间长，应该得到更多。

简单的解决方案

每次收益进来以后，根据比例，直接计算每个人的所得，然后直接转账分配收益。

这个过程一般通过异步任务完成。但是以太坊上，写入数据（改变状态）的成本很高，也就是计算复杂度十分敏感，上面的做法，每一次分配的复杂度是 O(N)，其中 N 是质押者的数量，因为每次收益都要更新所有人的账本。随着质押者越来越多，交易成本也越来越高，这个在以太坊上难以接受。

一种 O(1) 复杂度的更新算法

大概的想法是这样，首先需要改变一下体验：

• 可以在每次收益进来的时候更改某些变量就可以随时计算出质押者的总收益 O(1)

• 质押者需要通过一个提取操作来最终获得收益，合约在每次提取的时候记录下 O(1)

算法如下

设

T 当前的总股数
V 当前每一股可以获得的收益
stake[X] 质押者 X 质押的数量
snapshot[X] 质押者 X 上一次操作时候的 V 值

步骤

1. 区块1，A 和 B 分别质押 300 股和 100 股

T = 400, V = $0
stake[A] = 300, stake[B] = 100
snapshot[A] = snapshot[B] = 0

2. 区块2，进来一笔收益 $100

T = 400,V = $100 / 400 = $0.25
stake[A] = 300, stake[B] = 100
snapshot[A] = snapshot[B] = 0

3. 区块3，B 追加 200 股

B 自动提取收益 stake[B] * V = 100 * 0.25 =$25T = 600, V = $0.25
stake[A] = 300,stake[B] = 300
snapshot[A] = 0,snapshot[B] = $0.25
// 只需更新 B 自己的状态, V 不变, 因为 B 已经提取了截止当前的收益

4. 区块4，进来一笔收益 $200

T = 600,V = $0.25 + $200 / 600 = $0.5833         V = V' + revenue / T
stake[A] = 300, stake[B] = 300
snapshot[A] = 0, snapshot[B] = 0.25

5. 来看看双方剩余待提取的收益

A: stake[A] * V = 300 * $0.5833 =$175
B: stake[B] *(V - snapshot[B]) = 300 * ($0.5833 - $0.25) = $100
// B 已提取 $25, 还可以提取 $100, 应得总收益是 $125

其他类似的问题

固定速度收益（挖矿）

如果收益不是资金进来的时候产生，而是以一个固定的速度持续产生，比如每个区块产生 $1 收益。可以用前面类似的算法来实现，但是 V 是一个随着时间增长的变量。

存款利息（复利）

如果质押的不是“股”而是 $，也就是质押的资产和收益的资产是一样的，在固定速度收益的场景下，持续产生收益的过程中，质押的数量也会越来越大。用前面类似的算法可以算出复利所得。

复利是天然简单的计息方式，但不直观。而单利是在特定时间窗口里折算出来的利息，更直观。

如果资金池通过单利来控制收益，资金池就会有资金结余，作为储备金，通过单利调节需求。

闪电贷实现

合约代码

function flashLoan(
    address receiverAddress,    // 接收资产的地址
    address asset,              // 资产名称
    uint256 amount,             // 数量
    bytes calldata params       // 自定义执行参数
) {

    // 1. 记录余额, 把指定数量的资产转给 receiverAddress
    uint256 balance = IERC20(asset).balanceOf(this);
    IERC20(asset).transferTo(receiverAddress, amount);

    // 2. 执行自定义方法
    IFlashLoanReceiver receiver = IFlashLoanReceiver(receiverAddress);
    // receiver.executeOperation 里可以做任何事情, 但最后需要把 asset 转回来
receiver.executeOperation(params);

    // 3. 检查 receiverAddress 是否在步骤 2 执行完以后归还资产并支付手续费 0.3%
    require(IERC20(asset).balanceOf(this) >= balance * 1003 / 1000);

    // 4. 如果上面的检查失败, 则 flashLoan 执行失败, 交易回滚（状态变更不会被记录）

}

提供闪电贷的资金哪里来

来自于 DeFi 项目

1. Decentralized Exchanges 去中心化交易所: Uniswap, Sushiswap, etc.

2. Lending Protocols 借贷协议: Aave, Compound, etc.

3. Stablecoins 稳定币协议: MakerDAO

不是所有 DeFi 项目都支持闪电贷，最早支持的项目是 Aave，后来还有其他链的 DeFi 项目比如 BSC 的 DODO 也添加了这个功能，比较早期或者规模较小的链上存在大量的套利空间，需求更多，同时闪电贷可以给 DeFi 项目带来手续费收益。

去中心化交易所 DEX

交易所 DEX 原理 (Uniswap)

借贷协议

银行 Lending 原理 (Aave)

稳定币

未完待续

原文链接 https://consensys.net/blog/research-development/modelling-stateless-ethereum-a-journey-into-the-unknown/

什么是无状态以太坊 ?

在回答这个问题之前，我们先了解下什么是”状态“，以及为什么我们要摆脱“状态”。

以太坊世界的状态是指所有的以太坊钱包地址、它们的余额、部署的智能合约和相关的存储。由于新的钱包不断产生，新的智能合约也不断被部署，所以从当前的设计上来看，以太坊的状态数据会越来越多，无限增长。

而无限的状态增长意味着：新增一个新的以太坊全节点需要更长的时间，并且运行全节点需要更多的存储空间。带来的后果就是以太坊矿工运行节点的成本也会不断增加，运行节点挖矿的吸引力也就随之降低，最终以太坊难以规模化，节点越少越不安全。

Vitalik Buterin 早在 2017 年就认识到了这些问题，当时他首次提出了无状态以太坊的概念，主要目的是：通过缓解无限的状态增长，使以太坊规模化。

无状态（Stateless）的概念

确切的说，“无状态”并不真正意味着“没有状态”。就像我们说无状态请求或无状态服务的时候，说的“状态”都是指你产生了对别人有影响的东西，也就是说，你把维护和存储以太坊状态的责任转交给了网络中的另一个参与者，以太坊上现在就是这样的。

因此，“无状态”其实上是一个错误的说法，无状态的以太坊客户端并不是完全没有状态的，而是各自选择了自己想要维持的状态，大家相互之间不影响。

一个无状态的以太坊客户端应该能够选择它想维护的信息，并且持续更新它们。反之，它可以完全无视那些自己不感兴趣的状态，并且不会产生任何影响。

关于区块“见证”

实现无状态以太坊的关键机制是区块”见证“（见证是个密码学概念，后面会慢慢解释），在无状态以太坊中，当客户端从矿工那里收到经过验证的区块时，他们也将收到这个区块相应的见证。区块的“见证”由执行该区块所含交易所需的所有数据组成。

但如果是这样，区块的数据量就会很大，将有更多的数据包在网络上传递，这就得评估对网络产生的影响，以确保以太坊生态系统在这种变化的环境中继续安全有效地运行。

这就引出了一个关键问题：无状态以太坊是否可行？

要回答这个问题，我们首先来建个模。

关于建模

建模是预测未来的常用手段，也可以用来探索，当我们想改变一个稳定运行的环境时，可能出现的无意的连锁反应。

无状态的以太坊就是这么一种场景。目前的以太坊网络已经存在而且运行良好，我们可以不断获得大量靠谱的数据，然后也很了解它的运作方式。如果我们引入无状态以太坊，可能会打破现有的平衡，首先得搞清楚：改变后的系统将如何调整和运作。如果我们开始实施无状态以太坊，很多不确定的事情得提前了解清楚，比如，最坏和最好的情况是什么？

建模可以帮助我们捕捉已知的东西，也有利于我们对未知的探索。基于我们对已知事物的了解，我们可以对我们正在研究的系统中的关键过程(process)和相互作用进行建模。对于未知的东西，我们用概率来表示模型中的不确定性。我们也会咨询那些对当前系统或问题有深入了解的专家，把他们的知识在模型中表达出来，这个很重要，有助于定位哪些过程(process)可能会受到系统变化的影响。

另外，建模也要平衡必要的系统细节和模型的紧凑性。虽然模型里最好包含所有可能影响结果的因素，或包含构成一个复杂过程的每一部分。但有些因素对系统的整体行为的贡献有限，反而还增加了模型的混乱程度。因此，我们要咨询专家，确定最能体现系统行为的“关键因素”，或其中的一部分。然后，其实那些不太相关的细节所表现的行为往往已经包含在关键因素的行为中了。

后面的系列将一步步来解释我们为无状态以太坊建模的方法。

参考资料

V. Buterin, “The Stateless Ethereum Concept”, 2017.B. Edgington, “What’s New in Eth2 – 10 January 2020”, 2020.G.I. Hotchkiss, “The 1.x Files: The State of Stateless Ethereum”, 2019.

#205

!0bf413340f0db024d3621259bc4140f9defad64ac78725d90f569a131be561f4

#268

!413a4990d7788b4449ea4e322b03e582531ec699d4c08440a1c60f1031e425e6

为什么我们可以在文章里提前公开图片？

然后有人问我一个问题，因为现在 HippyGhosts 还没有开图，我是不是会根据 mint 的结果来对应分配稀有款，或者之后给某些地址开出特定的图？答案是不会。

因为所有 9999 个 HippyGhosts 的图片已经冻结，这些图片的 base64 编码进行 keccak256 哈希以后构成的 MerkleTree 已经提前生成并记录在了合约里。每当一个 Ghost 的图片公开了以后，任何人都可以验证图片的完整性和图片的位置（也就是 Token ID）。

所以，我这会儿说的稀有款，它的编号 #205 和它的图片内容，都是提前已经确定好的。而“提前确定好的图片内容和位置“这两个事实，任何人都可以在图片公开以后验证。

同理，未来开图以后，看到的每一张图，它的顺序和内容也已经提前确定，并且可以被验证，这样以确保未来开图过程的公平。

那比如说，如何验证上面这张图？

现在这个 HippyGhosts #205 提前开图，我需要验证，这个图不是我现在临时决定的，并且这个图，我在 mint 之前，已经把它排在了 9999 个图中的第 205 位。

首先，在新窗口打开这个图片，浏览器中可以看到这个图片的 base64 编码，这就是图片的数据，对这个数据进行 keccak256 哈希后，得到

0bf413340f0db024d3621259bc4140f9defad64ac78725d90f569a131be561f4

然后在 HippyGhosts 的 MerkleTree 中寻找这一串哈希，位置为第 205 个叶子结点

https://api.hippyghosts.io/~/storage/merkletree

因为这个 MerkleTree 是在 HippyGhosts 自己的服务器上的，接下来就是确认它的数据是否可信，首先从 MerkleTree 的第一行获得 MerkleRoot

58d247a687ef48f010e2e6107a04d575787163cfb0d70543c421a5001e9f5aab

同时这个 MerkleRoot 数据也存在于 HippyGhosts 的 Renderer 合约源代码中

https://etherscan.io/address/0x856bd414d7c4718f844795b30510af2f5fee2ee1#code#F9#L10

这里要提起一个叫做隔离见证的东西，意思就是：就是一个结果的出现，需要依赖于一个事实的发生。在这个例子中，结果就是 MerkleRoot，事实就是所有图片的内容和顺序。

58d247a687ef48f010e2e6107a04d575787163cfb0d70543c421a5001e9f5aab

现在我们在合约代码中发现了结果（即MerkleRoot），这个结果是在合约部署的时候就出现了，自然就是在 mint 之前，那么，所有图片的内容和顺序这个事实也一定发生在合约部署之前。这是因为根据 MerkleTree 的特点，如果图片的内容和顺序不同，MerkleRoot 也会变化，那么对于一个确定的 MerkleRoot，就一定有确定的图片内容和顺序。

讲完了，当然这不是公平开图的最有思路，只是我自己比较喜欢用 MerkleTree 来处理它。如果在公平开图这个事情上有其他想法，欢迎一起交流。

🔗 https://muselink.cc/web3nomad

直到 2021 年 8 月，受 Loot 启发，这群人里独立出三人，准备远离赛博朋克，远离金玉其外的喊单，远离浮夸的重复造车，拥抱链原生。在这个荒蛮的加密领域里，寻找更简单的目标和更硬核的 Web3 人类。接着，给靓图收藏家 Stani 的一次毫无意义的 pitch 准备中，我们组建了现在 HippyGhosts 小组的原型：Nomads。

那时候，ss，我，cafi，我们仨进入这个领域才差不多一年，因为下文《谁是 HippyGhosts》里说的一些稀奇古怪的缘由，有了第一个像素的嬉皮鬼，ss 称呼他 a hippy ghost。我们用优化的哈夫曼算法来压缩字符串，我们尝试着用点阵来编码它，以及用多份合约渲染来相互配合；尽管乐在其中，但32位的颜色编码和大量的压缩计算和字符串处理让我们对天价的 gas 开销心生畏惧。

https://mirror.xyz/cafi.eth/HAkCSuHQR85IbxyF1t4ImYaDIxg1k1GrAmnAV9TP2hg

就这样磨磨蹭蹭来到 2022 年，Discord 的人数已经数千，我们决定做点事情回馈社区，正儿八经的把 HippyGhosts 做成头像 NFT 发布出来。以此作为过去一年的总结，感谢过去一年帮助和影响过我们的人。并为未来新的目标加一些“噱头”，断了我们离开 Web3 的后路，把自己一路加密到底。

终于，随着 5 月到来，HippyGhosts 完成一次全量重绘，历经 3 个月，我们重新修订官网和合约。因为没有过硬的推广技能，我们决定躺平，设计新的发行方式，9999 个 ghosts 将在未来大半年里（~1.5M个区块）逐渐诞生，一起迎接以太坊2.0的合并。

https://hippyghosts.io/

🌈 言归正传。

发行方式

9999 个 HippyGhosts 包含 135 个稀有款，无序散落在所有 ghosts 之中；和 200 个 1/1 Ghost，留给团队自己，及赠送给这一年来帮助和影响过我们的人。

阶段一，社区铸造 Community Sale Mint

社区铸造（mint）阶段一共释放 1300 个 ghosts，包含 20 个稀有款，价格为 0.08ETH。这个阶段，我们保留了对发行节奏的控制，由 HippyGhosts 团队通过 Raffle、白名单活动和项目合作三种形式发放铸造凭证（白名单）。

获得白名单的人们可以参与社区铸造，具体信息建议关注 HippyGhosts 的 Twitter 和网站：

👉 twitter.com/hippyghosts

👉 hippyghosts.io

阶段二，公开铸造 Public Sale Mint

公开铸造阶段一共释放 8499 个 ghosts，包含 115 个稀有款。所有 ghosts 在设置了合约中的 publicMintStartBlock 以后，立即开始进行阶段性的荷兰拍卖，大约历时半年。

荷兰拍卖

公开铸造阶段一旦开始，就无法停止和修改，接下来：

• 每过 40K 个区块，就会有 300 个的 ghosts 进入可以铸造的状态。80K 个区块以后，就会释放 600 个，120K 就是 900，以此类推，直到 8499 个 ghosts 都释放结束。

• 每一批释放的 300 个 ghosts 价格都从 0.24ETH 开始，每隔 40K 个区块下降 0.04ETH，直至 0.08ETH。

• 铸造过程会优先选择最早释放出来的那一批 ghosts

所以，如果每个周期内释放的 300 个都被铸造完，价格会始终保持在 0.24ETH；如果有剩余，也就是铸造的速度跟不上释放的速度，价格会越来越低；随着铸造的数量逐渐跟上释放的数量，价格会逐渐上涨。

🤔 你可以等待价格下跌以后再来买，但有限的数量可能会被别人捷足先登。

最后，特别鸣谢官网上已经露脸的那些 “Hippy Ghosts” 们 👻 hippyghosts.io/ghosts

Let’s Hippy！

话说回来，我对于 crypto 和 Web3 依然长期乐观，自从 DeFi 构建起以太坊的金融规则以后，社区和组织的发展有了更好的基础设施来支撑。建议身边的 builder 们，有想法的可以做些贡献，没想法的可以默默支持，乌烟瘴气中给他们点时间，让这些好的社区在灰烬中崛起。

讲一些有趣的故事吧。

半个月前，一群年轻人成立了一个去中心化自治组织（Decentralized Autonomous Organization, DAO），叫做 ConsitutionDAO，他们筹集了超过四千万美元（$40M）用于购买一份罕见的美国宪法第一版的副本，这是现存的 13 份副本之一，将在苏富比（Sotheby’s）拍卖行进行拍卖。

他们是一群怎样的人呢？

其中年纪最小的贡献者，是个西班牙的程序员，只有 19 岁，标志性的紫色头发，思维敏捷，十分特别。整个项目核心成员基本是95后，他们代表着目前 Web3 社区贡献者的典型画像：

Z世代，社会经验少，熟悉区块链，热爱贡献，信任其他贡献者，不喜欢被管理

我觉得挺神奇的，自从企业开始和 MakerDAO 合作，金融机构开始和 Aave 合作以后，这是苏富比拍卖行第一次和 DAO 合作。因为参与竞标者都必须满足苏富比的 KYC 要求，只能以有限责任公司或是以博物馆身份参加，DAO 是新型组织，核心成员需要配合着整理一系列材料，一周都没怎么睡觉。并且这也是第一次苏富比接受 ETH 作为支付方式，要知道早前拍卖 CryptoPunk 的时候依然不支持 ETH。

而这群有趣的年轻人做到了，背后没有任何投资人，核心成员仅仅是一群志同道合者，各自在自己的社交圈传播。他们在主页上说明：这个 DAO 发行的代币 $PEOPLE 没有投资价值，只能证明出资者在 DAO 中的投票权，用来共同决定竞拍成功以后这份宪法副本的去向。参与者大概有12000人，尽管最近有上百美元的矿工费，大多数人依然毫不犹豫的参与众筹。有人说这次众筹中消耗的矿工费已经够买下本次拍卖前面的两三幅画作了，当然这只是玩笑，矿工费是 DAO 降低信任成本的基础。

这个众筹有风险吗？当然有。首先是合约漏洞，还来不及审计；其次是项目方跑路，毕竟这是在交易链下的实体。

后来请教了一位不愿意透露姓名的老师。他说每个年代的人都有自己的组织方式，所以目前的 DAO 和几百年前农民起义其实也没有差别。随着社会的进化，设施的进化，知识的进化，组织形式也会进化，没有对错和好坏，只要合理，人们就会选择它。

我想了下几个 DAO 的历程，不管是 MakerDAO，还是 Aave，抑或是 ConstitutionDAO，都经历了三个阶段

1. 为社区贡献积累声望

2. 建立一个组织，一呼百应

3. 和传统行业进行合作或对抗

区块链上没有信任成本，这一切自然更容易发生。当然 DAO 也不是完美的，就如比如三国或者再早点的秦末，估计也有很多人打着起义的旗号骗钱，和现在那些自利自大的傻逼 crypto 项目一样，最后倒霉的都是平民。

成功的组织，不光要顺应时代，还要做好人。

他们成功了吗？

没有😂成功。整个过程非常有趣，竞标于北京时间11月19日早上七点半在苏富比（Sotheby’s）拍卖行如期举行，直播在 YouTube 上，而直播聊天室明显已经被 Web3 社区霸占，大家基本都是第一次看艺术品的拍卖直播，尽管很新鲜，也完全没心思看前面几幅画作的拍卖，直到三十分钟以后宪法副本正式登场。

竞拍的有两方，一方的拍卖师是学术范的 David Gerard，另一方的拍卖师是贵族范的 Brooke Lampley。直到拍卖结束也还没有人知道代表 ConstitutionDAO 的是谁。过程中看到 Brooke 全程谈笑风生，气质不凡，丰富的肢体动作和表情，让对方的 David 毫无存在感，大家希望她是己方，又担心她是对手。所有人沉浸在双方20多次竞价的过程里，一边分享现场的趣图，一边毫无线索的寻找着答案。

直播聊天室中不乏一些知情人的消息，但因为推广工作不到位，所有的消息也都没有被证实，哪怕是平时和 DAO 成员接触紧密的一些哥们儿也都完全心里没谱。毕竟 ConstitutionDAO 的成员还是程序员和设计师为主，又很年轻，市场经验不多，拍卖前又忙于材料准备和测试，背后也没有大机构的协助，很多引导没有做好也可以理解。

在 Brooke 为自己的神秘买家拍下的一刻，大家的第一反应是移步 Twitter 问同一个问题：“我们赢了吗？”然后经历长达 30 分钟的情感过山车，先是以为自己成功了，欢呼雀跃并且开始计划 DAO 的下一个竞拍目标，甚至有人为 DAO 想好了更详细的“购物计划”。然后有人说被神秘买家截胡，接着再被大规模辟谣，直到 ConstitutionDAO 自己发帖证实竞拍没有成功（他们自己更紧张吧，过程中都没人来发个帖子啥的）。

本以为大家会不欢而散，戏剧性的事情却紧接而来。Brooke Lampley 在拍卖圈早有名气，但 ConstitutionDAO 的竞标失败反而让她在 Web3 社区彻底火了，大家觉得这一次失败原因是没有请 Brooke 来做拍卖师，为她做了有爱的偶像视频 (感兴趣的点这里👉brooke fan cam)，呼吁 DAO 的成员下次购物就找 Brooke。

有人嘲讽说 Brooke 只接曼哈顿八街的富人单，怎么可能来理我们这群坐地铁的泽西青年。而事实上 Brooke 已经连夜开了自己的实名 Twitter 账号来回应突如其来的粉丝，次日发帖表达自己对 ConstitutionDAO 的关注，并且暗示合作意愿，一瞬间涨粉上千，这谁能不爱？

所以赢家是出圈的 Brooke 吧。

下面讲个不一样的。

推荐个 app，叫做 Sonar

我是一个月前发现的，他们团队在里斯本参加 LisCon，而被一些人熟知。

简单说下它的功能，这是一个用 Emoji 和声音构建的“元宇宙”（我不喜欢用这个词，但为了显得自己无知，我就这么叫吧）。

1. 每个人在里面是一个 dot，它可以在 server 中移动。

2. 每个人都可以创建无数个自己的 server (也叫房间)，它是一个有边界的空白平面，相互间完全独立。

3. dot 可以扔 Emoji 在自己创建的 server 里构建一个平面世界，比如墙、树木、生活用品、文字板、电器，色块等等。

4. 其他 dot 可以在 server 里的空白区域随意移动，所以 Emoji 围起来的封闭的区域无法进出。

5. 直接说话可以让 dot 发声，靠近 TA 的 dot 才可以听到，所以说话的时候大家会围成一团；可以鸣喇叭，但会被骂死。

6. 一些特殊的 Emoji 有专门用途，只能自己发掘，比如扩音器可以让更远的 dot 听到自己说话，比如喷泉有水声。

7. 游戏里面有金币，目前会随机出现在 dot 比较多的 server 里，谁抢到就归谁，app 里没有实现任何金币的用途。

Sonar 的产品设计有些诡异，似乎是故意设计了一些反常的交互，也没任何帮助文档，很多功能需要在 server 里找到团队成员问他们，可能是为了营造一种在空白二维世界里摸索的感觉。

排除少量只会口吐芬芳和大吵大闹的垃圾 dot，整个社区还很和谐。目前内部已经形成了多个微型的生活和关系圈，项目团队的日常会议，观点分享，V.I.D 的投票，甚至一些在线 BUG 调试，都会发生在有特定场景的 server 里。

上个月 Sonar 发了一个 NFT 叫做 Moji，虽然地板价现在跌的很惨，但依然挺好玩的。也正是因为项目方没有在社交网络上哗众取众，大家反而更珍惜，拥有 Moji 的人在 app 里可以为自己的 dot 换上动态 3D 头像，十分拉风。

目前 Sonar 的 app 用户还不多，目测还只有10K，我是早期的 Moji 买家，所以 TokenID 很小，比较好找，有兴趣的人可以在里面自己搜寻。项目创始团队 7 人，其他社区运营者和开发人员基本都是在校实习生。

Sonar 对自己的定位是 Web3 社区，主要推广手段是人传人，目前通过搜索引擎不好找，Discord 不算活跃，团队运营能力也还一般，里面美国人为主，少量欧洲人，这里贴个链接吧 https://sonar.app/

最后讲个正经点的，简单说说。

关于科学创新

上周一个大学同学过来天潼路找我，也就是我文中经常出现的门飞，他送了我两本书，《科学革命的结构》和《科学发现的逻辑》，我摸了一下他的额头，以为他发烧了，怎么突然对科学哲学感兴趣起来，还各买两本，送我一本。他摸了摸自己的后脑勺，笑呵呵提醒我前段日子我们参与的一个话题。

其实是这样的，就一个小事儿。

现在像样的 Web3 项目已经不少，大家喜欢说自己在范式转移，也就是 paradigm shift，然后有人就发现些问题：所有 NFT 和 xxFi 这类场景，基本都是可以描述成，”xxx的web3版本“或者”xxx的crypto版本“，这种感觉很挫。

而事实上每一次技术的跃迁都有这个现象，做新东西不等于创新。范式转移要关注 unique feature。

举个例子

十几年前，高中那会儿，学校在实行数字化，每个教室配备了投影仪和电脑。学校一般会给老师一年时间进行教学计划改造。大部分老师选择把书本和手写讲义扫描成图片，然后直接投影或者传文件给学生；只有少数老师会用数字软件重新做课件，然后才能做出互动性更强的讲义。后者就是现在大部分在线课程的形式。

互动性就是一个线上课程的 unique feature，它是不可逆的。这个课堂数字化升级的过程中，大多数时候我们只是在做简单的搬运，但搬运只是时间问题，除了互动性，其他事情根本不重要。

再想想最近半年的元宇宙概念，那些试图复制物理世界的炫酷引擎，还不如 Roblox 这样的开放世界或者 Minecraft 这样的像素世界来的有意义，哪怕是简单至 Sonar 这样的平面世界，目前都有能力对大部分的元宇宙项目进行降维打击。

回过头来看 crypto 领域，想想闪电贷（无抵押无风险贷款）就是 DeFi 的 unique feature，所以 DeFi 才会变的无可替代，他对传统金融的转变是不可逆的。我们无法用“xxx的crypto版本”来描绘闪电贷，这就让 DeFi 成为区块链上又一个屹立不倒的创新。而很多 DeFi 项目被人利用闪电贷攻击也是因为对它没有充分重视或者了解导致的。

看看，把赌博搬到以太坊上不叫创新，把期货搬到以太坊上也不叫创新。我们日常构建归构建，也需要时不时想想，自己真的在创新吗，还是只是历史潮流上的搬运工？

所以 web3 或者 crypto 领域还有其他 unique feature 吗？

也许有，比如 zk（零知识证明）等，但都还不成熟，其他的说实话还没找到。所以愿景可以有，信仰可以有，但真要创新，颠覆传统，还需要再努力一把。

创新在所有领域都一样的，多看看书吧。

// We’re All Gonna Make It

I. Upgrade the system Keeping the system in an updated state is the best solution for this kind of problem, but if it is inconvenient to do a complete upgrade, please focus on upgrading openssl, gnutls and ca-certificates.

CentOS / RHEL

yum upgrade openssl gnutls ca-certificates

Ubuntu / Debian

apt upgrade openssl libgnutls30 ca-certificates

This solution is available for the following platforms.

Windows >= XP SP3
macOS >= 10.12.1
iOS >= 10
Android >= 7.1.1
Mozilla Firefox >= 50.0
Ubuntu >= xenial / 16.04
Debian >= jessie / 8
Java 8 >= 8u141
Java 7 >= 7u151
NSS >= 3.26

II. Disabling expired certificates manually If the system no longer provides updates, or it is inconvenient to update the system, you can manually disable the expired certificate, the specific operation scheme is as follows.

Linux platform

Open and edit the /etc/ca-certificates.conf file, and add a ! (exclamation point, English, half-word) to disable the certificate so that it reads !mozilla/DST_Root_CA_X3.crt. After editing, run the update-ca-certificates command to update the system's certificate chain.

On CentOS 7 and later

you need to execute the following command: cp /etc/pki/ca-trust/extracted/cadir/DST_Root_CA_X3.pem /etc/pki/ca-trust/source/blacklist update-ca-trust

Windows Platform

Use the shortcut Win + r and type certmgr.msc to open the system's certificate manager, search for DST ROOT CA X3 and delete the relevant certificate and reboot.

Java Platform

Execute the following command: sudo keytool -delete -alias dstrootcax3 -cacerts -storepass 'changeit'

MakerDAO有条件使用有史以来最强大的工具来对抗有史以来最具挑战性的问题–所需的只是一个社区的意愿。

随着Maker基金会的解散，我有机会对项目的未来进行反思。我们已经取得了令人难以置信的成就，今天的DAO正因大量的治理和协议活动而热闹非凡。但我相信，仍有一个根本性的机会来重新思考我们如何对待Maker治理，使其更有活力，更有针对性，更分散。

为了真正发挥其潜力，Maker需要成为一个目的驱动的DAO。最重要的是，我们的去中心化社区需要一个引人注目的愿景。一个强大到足以作为北极星的愿景，减少分心，使社区专注于执行和结果。同时，这个愿景也必须通过向世界展示创客和去中心化金融的变革潜力来吸引非加密货币原住民。

从一开始，Maker的目标就一直是利用区块链技术的力量为人们和社会提供实际利益。到目前为止，我们的方法是直接关注金融包容性，我们已经在阿根廷的Dai采用方面取得了成功，DeFi的寒武纪爆炸证明了人们对不受约束地获得金融有很多需求。

一路走来，我们也了解到，Maker作为后端基础设施协议确实很出色，同时也体会到DAO在物理连接最后一公里和提供那种用户体验方面的局限性，这将真正导致直接、大规模的金融包容性。

举个例子，今天从ETH产生的所有Dai中几乎有一半来自仅仅5个金库，而这5个金库中至少有一些是CeFi公司，它们向零售或机构用户提供注重用户体验和合规性的规范贷款。同样，现实世界资产中存在的最大和最可扩展的机会依赖于将大规模的信贷额度扩展到受监管的机构，然后可以将其输送到更广泛的经济中。法国兴业银行–世界上最大的银行之一–最近提出的治理建议，证明了Maker作为一个新的金融后台，有可能影响金融的最深层。

这种对Maker今天真正开始发挥的作用的认识，引导我们走向它的真正潜力，不仅是作为一个以终端用户为中心的DeFi协议，而且是作为一个金融基础设施，可以帮助重塑使用它的其他金融机构。从这个位置出发，Maker可以帮助修复全球金融系统本身的破损核心，以及推动其走向自身灭亡的不良激励机制。

气候变化和工业社会的崩溃

每个人都知道气候变化是糟糕的–残酷的现实每天都在以洪水、野火和影响世界各地人民的无数其他灾难的形式变得更加明显。IPCC已经为我们提供了无可辩驳的证据28，即一切照旧只能再持续10年，这一信息已经被世界各地的科学家反复分享。

最大的问题是气候临界点，一旦过了这个临界点，将通过正反馈循环锁定不可逆转的破坏性变化……而我们几乎就在那里。一个例子是永久冻土的融化，这是一个由全球变暖引起的事件，然后它本身以不断加速的速度排放出大量的温室气体，相当于整个工业化国家。

还有许多其他类似的现象。野火、森林枯萎和冰雪融化。所有这些都是破坏性的正反馈循环，加速了全球变暖的趋势，并使其不可逆转。在实践中，为了防止它成为灾难性的和无法缓解的，人类必须根据下图积极地加速减排。

可悲的是，迫切需要的那种全球协调和系统性的变化根本没有发生–尽管所有的自然灾害都应该引起支持气候行动的警报。中国和美国–世界上最大的排放国–仍在加速使用化石燃料，大多数发展中国家继续从根本上依赖扩大化石燃料能力来推动增长和帮助其人民摆脱贫困。社会继续争论气候变化威胁的可信度，全球金融体系仍在优先考虑破坏性的短期收益–通常对生态环境有非常负面的影响–而不是长期的、更可持续的利润。

现实情况是，我们所知的地球将发生巨大的、不可逆转的变化，而这种变化将产生非常真实的经济和社会影响。当大片的农业用地变得贫瘠，或者当数十亿人因为他们的家园变得荒凉而流离失所，会发生什么？这将对现有政府造成难以想象的压力，并迫使地缘政治力量发生巨大的转变。

这张地图有助于描绘出如果不采取任何措施，情况会变得多么糟糕。

一些更精确的数据。

• 气候变化跟踪最坏情况 – YouTube 17，显示了我们目前如何跟踪或超过最坏情况的排放，称为 “RCP8.5″，这个模型最初被设计为不切实际的坏情况，而不是在现实中可能发生的事情。

• 气候变化有可能将全球大部分粮食生产和人口中心推向前所未有的状况–NASA/ADS 2显示，如果我们继续跟踪RCP 8.5，那么仅在未来50年，1/5的可耕地将变得无法生产，1/4的人口中心将变得不适合居住。

• 气候变化是最广泛研究的科学课题之一。我鼓励你自己做更多的研究，因为每天都有大量的材料在发表。

当气候变化在80年代首次变得明显时，社会的主要反应是把罐子踢到路上，而不是把它当作系统性的威胁和存在的风险，甚至今天大多数政治家会转移问题，并指出 “技术解决方案”，由于人类的聪明才智和复原力，肯定会出现解决所有的问题。

这种银弹的幻想招致大众对我们的经济和生活方式将从根本上改变的事实感到自满–没有多少创新可以在短短几十年内改变热力学规律的影响。

在这个后期阶段，仍然只有一种选择可以上升到应对这种存在的风险。我们必须利用我们人类最强大和最古老的工具：金钱的力量。

我们能做什么？

Maker和DeFi与气候变化有什么关系？气候问题的核心实际上是一个金融和资本分配问题。其根本原因是全球经济没有能力为短期以外的事情做计划，也没有能力处理其集中和不透明的基础设施中存在的根深蒂固的腐败–这就是为什么它允许自己达到破坏世界的污染水平，并继续为其输送数十亿的资本，尽管这是一个无可辩驳的事实，这将是其自身的毁灭。

DeFi的存在正是为了处理这种类型的问题。

比特币开启了区块链革命，它的诞生是对2007年金融危机的反应，这场危机与气候危机非常相似–只是规模小得多。在这两种情况下，我们都看到了系统的根本无力，无法超越非常短期的眼光。派对必须继续下去，没有人有动力去关心音乐停止后会发生什么。

透明度、利益相关者治理、精心设计的激励机制、去中心化以及将未来长期价值转化为当前现金流的能力是世界所需要的，而这正是Maker、Dai和DeFi能够提供的。

通过利用区块链的基本特征，我们可以开发出可验证的流程，以确保所有Maker的抵押品都是可持续的、与气候相关的资产，考虑到金融活动对环境的长期影响。这将释放出Dai作为世界迫切需要的协调工具–一种让人们和公司联合起来的方式，直接在问题的核心处创造出现实的影响。

这场危机是如此严重，而且处于如此晚期阶段，全球的反应需要不亚于 “全面战争”。全球经济的每个方面、每个公司的商业模式和每个人的生活都将发生变化，要么主动地通过合作来避免灾难，要么非自愿地承受不作为的后果。

一个古老的想法，其时代已经到来

可持续金融原则21主张用可持续和符合气候的抵押品支持Dai，这个概念最初是由创客社区在2018年首次创客治理投票中批准的，作为5项基金会原则的一部分。这不是一个新的概念，是一个古老的想法，其时代已经到来。

但是，这不是简单地确保一些RWA抵押品是在太阳能和风能资产中。为了认真地与我们所知的世界末日作斗争，Maker社区将不得不真正采用可持续金融的原则作为Maker文化和整个抵押品战略的核心。气候变化对金融稳定的影响至少在未来50年内会比其他任何事情都大。这个时间段，尤其是当前的十年，也是决定危机结果的关键机会窗口。

拥有一个重点突出的抵押品战略为制造商提供了巨大的好处。其中最明显的一点是通过定义一个具体的、有界限的范围来减少横向治理的复杂性和信息过载。它还允许采取积极主动的方法，让MKR持有者决定他们想要承担什么样的风险。

作为清洁资金倡议的一部分，我提出了一个基于核心投资组合的抵押品战略，其中包括 “旗舰抵押品 “的三要素。

可持续的抵押品

Maker和Dai可以协调可持续能源途径的全球建设，将数十亿美元的资金注入到建设太阳能农场、风力涡轮机、电池、充电站和其他具有成本效益的可再生能源解决方案以及其供应链、可持续资源开采和回收等项目的高级信贷职位。

仅仅是可再生能源还不足以实现可持续的经济，但这是一个Maker很好地处理的领域，因为它允许Maker治理公司在一个定义明确的市场内建立专门的风险评估专业知识，该市场是商品化和高度可扩展的。

今天，我们已经拥有了我们所需要的一切，通过使用社区多年来开发的基于受托人的真实世界资产模型，开始将我们的RWA风险暴露扩大到数千亿美元甚至更多，安全并完全符合金融监管的要求，该模型最近由6s资本推出，其他项目也在路上。

使得可再生能源和可持续抵押品非常适合Maker的一个独特优势是，如果采取适当的措施，通过强有力的营销和沟通以及像NFTs这样更奇特的方法来支持这些努力，那么为社会创造的积极外部因素可以被Maker以社会和政治资本的形式间接地资本化，从而减少政治不确定性。

有弹性的抵押品

其次，我们需要确保Dai背后的很大一部分抵押品是可以证明有弹性的抵押品，可以抵御气候变化的影响。

从积极的角度来看，为什么具有气候弹性的抵押品对创客的未来至关重要，这就是所谓的 “气候阿尔法 “的概念。今天，大多数资产的定价是错误的，因为气候风险没有被我们目前的资本市场考虑在内，因为–在这里重复我自己–这个系统无法考虑任何东西，而只是考虑非常短的时间。尽早抓住气候阿尔法可能是未来几十年最大的经济机会，因为事情变得多么糟糕–就像在2010年发现比特币，但在宏观层面。

无论我们或其他任何人在防止灾难性气候变化的事业上付出多少努力，我们都需要为一个部分全球崩溃的未来做好准备，在那里，许多国家和我们今天认为理所当然的整个地区将挣扎，因为肥沃的土地越来越少，极端天气事件造成严重破坏，社会紧张局势不断加剧。

还有一种真实的可能性是，这些无休止的经济震荡将导致美元和其他世界货币的恶性通货膨胀。在这种情况下，至关重要的是，即使其他货币未能做到这一点，Dai也要准备好气候弹性，这样，如果最坏的情况发生，Dai可以独立存在，实现其成为自由流通货币的潜力。

这种弹性可以通过将我们的长期RWA抵押品的大部分广泛分散到具有气候弹性和社会政治稳定的国家来实现。

这些地方将成为人类抵御气候变化影响的堡垒，并将为傣族提供它所需要的力量，即使在其他金融多米诺骨牌开始倒下的时候也能独立生存。将资本输送到具有气候复原力的国家也很重要，因为这些国家迫切需要发展，以便拥有粮食生产、住所和工作岗位，以容纳本世纪将移民到那里的气候难民，特别是如果气候变化不能得到很好的缓解。

在现有的少数具有气候复原力的国家和少数地缘政治独立的国家之间存在着偶然的重叠，这些国家的政治制度适合容纳大量的DeFi抵押品，而且监管风险最小。这两类重叠的国家可以被认为是 “超级国家”，它们应该为Dai提供大部分的现实世界的资产风险。

我所认为的超级国家的例子是。新西兰、加拿大、瑞士和英国。

去中心化的抵押品

最后，我们需要加倍巩固我们的去中心化抵押品的根基，特别是巩固我们对以太坊和ETH代币的基本承诺和依赖。以太坊区块链是为人类的协调和复原力而建立的。随着世界被迫应对气候变化的影响，以太坊将使金融市场能够协调缓解，即使在大规模灾害期间，它也将继续发挥作用。

重要的是，一旦完成从工作证明到股权证明的升级，以太坊将成为一个高度节能的区块链，ETH将成为比特币目前作为主要加密货币的可持续竞争者。

Maker的抵押品策略应该优先积累更多的ETH作为各种风险参数组合的抵押品，并积累更多由ETH衍生的抵押品，如LP代币或ETH抵押资产。直接持有ETH作为协议储备也可以被认为是Maker直接支持，并直接从其生态系统中受益的有力方法。

可能会有一个长期的过渡，即在DeFi中使用抵押的ETH，而Maker应该带头这样做，以防止它为其他人从Maker中获取ETH抵押品创造一个机会。由于抵押ETH需要第三方解决方案才能在DeFi中使用，因此Maker必须确保在DeFi中采用抵押ETH不会影响网络的去中心化和安全性。

另一个途径是依靠更广泛的DeFi生态系统，使用直接存款模块进入Aave、Compound和Yearn等平台，产生由分散的抵押品支持的Dai，这可以产生高收益，风险适中，并有助于提高Dai在DeFi生态系统中的市场地位和可用性。

稳定币是他们的食物

一个由可持续的和具有气候适应性的抵押品支持的Dai稳定币是一个强大的与气候相关的品牌。在一个受到气候危机严重挑战的世界，对Dai的需求将随着其抵押品的日益成功和相关性而增长。气候危机正开始对人们的身份和个人生活产生深刻的影响–特别是最年轻的几代人–最近的研究显示，超过一半的年轻人认为人类注定要失败，就是一个例子。56%的年轻人认为人类注定要失败。

你愿意使用清洁货币还是黑匣子货币？

随着Dai作为一个分散的、可持续的、有弹性的稳定币运行，由服务于上述每一个目的的资产支持，我们可以诚实和自信地称Dai为 “干净的钱”。

• 清洁是因为它在拯救地球，使其免受可持续抵押品的污染和负外部性方面发挥了作用。

• 清洁是因为它透明地适应气候变化和全球不稳定对其稳定性的长期影响，并提供分散和有弹性的抵押品。

• 清洁是因为它的治理和运营基础设施完全是公开运作的。

如果Maker能够现实地处理当下最相关的问题，我们可以催化一个新的时代，让人们认识到DeFi和区块链以及它们作为协调工具给社会带来的积极好处，同时也让人们注意到当前黑箱金融和货币模式的问题。

积极的短期增长的货币和金融体系是气候危机首先存在的核心原因，直到今天，美联储和欧洲央行的量化宽松计划并没有进行严格的环境、社会和治理（ESG）筛选，所以他们积极地将公众的资金直接输送到正在侵蚀其经济的化石燃料项目中，这是不可估量的。

通过向世界展示容易消化的证据，如用Dai用户的资金建造的实际风力涡轮机的图片，并辅以影响数据的科学报告，我们可以向人们证明，终于有一个现实的气候行动选择，个人可以帮助在问题的核心部分改变天平。

我们分配给可持续活动的资金越多，同时用大规模、高质量的营销和公关活动来支持它，我们的项目就能产生更多积极的气候影响。经过验证的结果将加强我们的品牌，导致更强大的采用漏斗和更便宜的资本成本。这将使有影响力的投资数量不断增加，为超增长的良性循环提供动力。

此外，Makers气候影响的主流意识和相关性也表现为政治弹性。如果创客能够带来不可否认的现实利益，那么在气候问题被列入政治议程的国家，监管风险将大大降低。

支持性抵押品

我已经提到了三类旗舰抵押品，但还有另外两类支持性抵押品需要考虑。

骨干抵押品

第一类是骨干抵押品。这一类别极为重要，因为它为傣族的稳定提供了支柱，特别是在傣族需求大量涌入的情况下。骨干抵押品的作用是拥有RWA基础设施，可以将大量资本吸收到安全、生产性和符合气候的资产中。这方面的完美工具是顶级ESG企业的公司债券，由位于超级国家的世界级受托人保护的信托机构持有。它们支付良好的收益率，它们加强了清洁资金的愿景，它们是安全的，它们可以在政治上安全的司法管辖区持有，并且它们可以有相对的流动性。

Maker将需要审查不同的ESG框架，以确定哪些框架是真正的，哪些只是洗绿，然后支持和改善真正的ESG标准–如果Dai发展得足够大，那么这将对ESG评级生态系统产生真正的积极影响。

流动性抵押品

除了骨干抵押品，我们还需要流动性抵押品，这主要采取流行的集中式稳定币的形式。为了使Maker不可能被单独列入黑名单，我们可以用 “DeFi盾” 来持有大部分的稳定币，比如向Aave和Compound的存款–这样我们也会从它们身上获得收益。

流动性抵押品的另一个潜在来源是全球金融机构和银行的流动性、安全和灵活的风险。法国兴业银行最近的代币化债券实验表明，银行可能愿意向Maker提供这种风险。直接面向机构的多样化风险敞口可以成为集中式稳定币风险敞口的有效平衡，并可以减少美国风险敞口。

Dai需要大量现成的稳定币，以确保对挂钩的信心不会被打破，至少需要1亿个，而且随着Dai流通总量的增加，需要更多。为了提高Dai的增长，并使其有可能与集中式稳定币竞争，PSM的点差应该减少到0，以便Dai成为一个适当的稳定币，没有问题的用户体验摩擦。不收取点差所损失的收入由Backbone抵押品所产生的回报来弥补。

解决USDC风险的决定性方案

我的估计是，如果我们现在调整行动方向，在一年内我们可以将30亿或更多的USDC风险分配到ESG企业债券中，由英国或其他超级国家的世界级托管人保护。我们还将能够对PSM中剩余的稳定币实施DeFi屏蔽–这些措施加在一起，几乎可以消除我们所有的美国风险。这将是一个巨大的成就，但考虑到Maker Governance今天所取得的一切成就和发展，实际上是比较容易执行的。

届时，我们将实现大规模的可持续发展转型，将数十亿资金导入ESG资产，同时协议将开始获得良好的收益，使DSR的重启开始进入超速增长阶段，我们将把风险分散到美国之外，因为美国目前的DeFi环境是不可预测的，并消除了用稳定币黑名单专门针对Maker的能力。

在我看来，在政治气候仍不确定的情况下，我们在美国的主要曝光类型应该是可持续的抵押品旗舰项目，这有助于展示DeFi能够提供的好处。

释放DeFi的超级力量

通过这个全面的计划，我们给Maker一个重要的目的。通过使Maker成为全球气候行动的工具，大大增加了权力下放和复原力，我们创造了动力、焦点和机会，使Dai和MKR的采用都有超常的增长。

让我们现实一点吧。为了使其产生世界迫切需要的影响，需要使参与其中的人非常有利可图。

我们需要将把资本分配给可持续性和发展复原力的公共利益与推动资本市场的个人动机结合起来。

传统上，这被认为是不可能的。你要么以非营利组织的方式拯救鲸鱼并最终消亡，要么通过无情的资本主义燃烧地球来获取利润。

但在我们今天生活的世界里，通过转向最近创新的DeFi “超级大国”，包括tokenomics、收益率耕作、代币发行和NFTs，使社会的长期价值与个人的短期价值保持一致是可能的。

这些都是经济机制，使其有可能在协调问题的积极结果下，将远在未来存在的抽象价值，转化为今天的硬现金流。如何实际落实这一点构成了清洁货币倡议的核心部分。

人马座的引擎

DeFi的最新浪潮是由创造性地使用先进的tokenomics和经济流来推动的。我为Maker提出的使清洁货币倡议成为可能并具有高度价值的核心要素是一个名为人马座引擎的代币经济学系统。这个系统将主要取代购买和燃烧作为协议剩余的目的地，并在MKR的价值模式上进行重大创新，借鉴DeFi夏季的过程中在DeFi生态系统学到的经验，以及随后的高级代币组学的繁荣。
人马座引擎的主要价值主张是，它给那些长期锁定其MKR的MKR持有者带来了切实的好处。这样做的主要动机是能够从人马座引擎的协议盈余中积累的Dai池中借款，用锁定的MKR作为抵押。这种借贷机会将以极具吸引力的利率（相当于DSR）和非常宽松的清算条款（至少1周的宽限期）进行。

用MKR抵押品借来的Dai不会在系统中造成尾部风险的积累，因为它是通过标准抵押品产生的 “正常 “Dai，只有在协议赚取了它的盈余后，才由人马座引擎借出。

人马座的名字来自于银河系中心的超大质量黑洞的位置，即人马座A*，它将整个银河系固定在一起，使其处于稳定的平衡状态。

该引擎旨在创造一个类似于 “大写黑洞 “的效果，该黑洞位于MakerDAO的中心，在强大的对准平衡中将其保持在一起。这方面的主要机制是通过直接将利益输送给活跃的社区成员，提供一个极强的投票和参与激励。同时，该模型不会泄漏资本，这意味着一旦资本越过 “事件视界 “并被输送到人马座引擎，它只能随着时间的推移而增长，尽管人马座的用户同时也能直接从中受益。

这有一些极其强大的长期效应，意味着一旦它达到足够大的规模，它将永远继续吸引大量的MKR，这些MKR将被锁定很长一段时间。

为了完成这个循环，并真正增强人马座引擎的潜力，我们需要执行大规模的MKR代币发行。今天，MKR的发行已经在发生，核心单位MKR的归属模块直接发行新的MKR来支付奖金。我认为我们有机会利用大规模MKR发行的力量来推动协议的增长，并使MKR持有人受益，就像DeFi夏季以来较新的DeFi协议所做的那样，同时也通过设置一个新的上限300万MKR来关闭潜在的无限发行的大门，防止核心单位发行新的MKR，而是引导他们利用协议库来支付任何MKR费用。

发行时间表应在未来50年内发生，通过为Maker成为全球气候行动的力量提供必要的燃料，加强并配合清洁货币的愿景。通过提供一个与项目目的相对应的长期时间表，我们可以设计一个冰河时代，这个概念意味着在一段时间内，治理可以不那么随意，而更注重耐心地执行一个长期的、可靠的计划，而不是新的创造性建议和支点。冰河时期的概念将在下文进一步描述。

为了进一步提高人马座引擎对创客治理的积极作用，锁定的MKR有一个2倍左右的投票权乘数（尽管这需要在社区里仔细辩论）。这意味着那些愿意致力于协议的长期成功，并在游戏中投入更多皮肤的投票者将有更大的发言权–这样一来，一个比大鲸鱼更投入的小持有人仍有可能仍有更有效的投票和更大的影响力。

结合大量新MKR的发行，人马座引擎有可能迎来一个新的时代，创客治理将看到大量新的血液涌入，允许大量新的中小型持有人进入生态系统，并在项目执行清洁货币愿景时对未来决策有很大的发言权。

通过MKR的发行来激励减少流通中的MKR

我们用人马座引擎发行的新MKR应该以各种方式分配，包括直接放在荷兰的拍卖会上，以及通过类似收益耕作的项目分配给分散的抵押品类型的用户，从而使MKR在整个市场上分配，并使现金涌入制造者协议。

这些现金随后被输送到人马座引擎的资本池中，以实现更多的MKR抵押的低成本借款。这将推动需求，以锁定大部分新发行的MKR，加上目前市场上大量的MKR。

由于短期内市场上的净效应可能会减少MKR，Maker Governance将有能力利用这些现金流，并将其中一部分转用于核心单位和其他开支，如建立长期储备，允许更雄心勃勃的资本部署战略，当涉及到对气候变化的影响时，可以发挥不同作用。

影响力NFTs：艺术家们新的可持续发展的原动力

为了尝试将代币机制与愿景直接联系在一起，并使MKR持有者能够从他们资助的积极外部因素中获取价值，人马座引擎将有一个基于影响力NFTs概念的NFT组件。其基本理念是，DAO实现的每个与愿景相一致的成功故事都被保存在NFT中，然后以代币加权抽奖的方式分配给人马座的用户。具体来说，这意味着将每个风力涡轮机、每个太阳能容量块、每个电池单元和其他可持续旗舰抵押项目的 “状态信号 “部分代币化，然后这些NFT将具有社会价值，因为社会重视那些为公共利益做出贡献的人。

这些NFTs可以利用Maker Oracle网络对影响的规模进行实时统计，如产生了多少清洁能源，抵消了多少碳–在某些情况下，可以进行创造性的解释，如显示种植的树木的等值数量和大小，以及其他创新的方式，试图使创造的社会影响和积极的外部性更加具体和可亲。

DAO可以使用适度的预算来资助新晋的NFT艺术家和开发者，在这些NFT中开发艺术作品和游戏化功能，以提高其价值，并增加基础旗舰抵押项目的独特性和历史意义及其气候影响。随着时间的推移，可以尝试许多不同的方法，所有这些方法的共同点是国家信托基金与现实世界的气候影响相联系。

影响力NFTs的价值越大，它们就越能帮助气候行动的事业，吸引更多的MKR进入人马座引擎参与抽奖活动。此外，对影响型NFTs的交易收取 “版税 “意味着如果它们被交易，部分价格可以被输送到慈善项目，如红树林重新造林–这些项目本身可以被用来创造稀有和独特的影响型NFTs。

除了仅仅创造具有相关艺术作品和统计数据的影响型NFT之外，还有一些更先进的趋势，今天存在的Maker可以采用。目前，至少有两个主要的 “基元 “出现在NFTs中，这两个基元都可以采用，并提供气候影响的旋转。

• 生成性的个人资料图片，如旨在孵化社会媒体社区的cryptopunks。

• 像Loot项目这样的游戏基本要素，建立了一个基本的构件，然后指望其他人在此基础上不断增加游戏化的维度，只要有一个参与的用户群。

随着时间的推移，可能会出现更多类似的趋势，Maker可以不断地旨在采用那些适合影响NFTs的趋势。

大多数艺术的灵感来自于自然和我们的物理世界，并与之紧密相连。吸引其他人与影响力NFTs合作并在其基础上进行建设，可以帮助催化许多新的艺术作品，并为艺术家提供一种方法，使他们的创作对应对气候变化产生直接的积极影响。

由MKR持有人提供，为MKR持有人服务

有了一个与世界人民的需求产生共鸣的强大愿景，再加上同样强大的tokenomics和加密货币超级大国的支持，一切都准备就绪，Maker将在未来几十年内以光速发展。

差不多了。

拼图的最后一块关键部分需要像愿景和代币经济学一样强大：执行和规范所有机制的内部官僚结构，并在超增长阶段和之后不断适应新世界。

只有可信地能够处理全球经济中如此大规模的增长和重要性，Maker才有可能让公众接受清洁货币的愿景，从而挖掘愿景的未来价值，以便在短期内推动增长并在未来几年内启动这一进程。

我们需要发展创客治理过程、文化和政治动态，使其在大规模的增长和很长的时间段内，在高度流动和不断变化的环境中，也能做到极为分散和高效。

我们还需要创客治理过程从根本上以MKR持有者为先，这样可以鼓励MKR持有者的基层活动水平比今天看到的要高得多。MKR持有者拥有投票权，是协议完整性的最终监护人，但只有当他们被授权并充分了解情况，作为分散社区的骨干时，这才行得通。人马座引擎已经在很大程度上实现了这一点，但创客治理的规则、框架和官僚机构也必须从这个角度进行根本性的设计。

创客治理必须发展成为由MKR持有者组成，为MKR持有者服务。

如果我们能够将 “清洁货币 “愿景的成功及其所能创造的公共利益与MKR选民的自我利益结合起来，那么就有可能维持一个稳定的、自我调节的治理平衡，最终由MKR持有者的个人自我利益来执行和推动–特别是那些被锁在射手座引擎中的人。

人马座的时代。下一代的创客治理

治理地震 “是指一个社区没有足够数量或质量的规范、规则或程序来处理其外部环境，因此它留下了一个空白，使新提案的比率很高，有可能通过并改变现状。那么，新提案的高比率本身就使快速变化以及对任何既定事物的批判性看法正常化–这意味着一个治理领域的快速变化会导致所有治理领域的快速变化。

治理冰期 “则相反，它是一种情况，即治理框架的质量和数量足以 “充分覆盖所有基础”，这意味着对于任何特定情况，已经有规范和规则可以很好地处理它们。在这种情况下，很少有必要提出新的突破性建议，即使提出了，也会受到严格的审查，对现状的改变也会缓慢而谨慎地进行，以确保造成最小的破坏。这就使最小的变化正常化并得到加强，创造了一个很少提出建议而且几乎总是被拒绝的环境。

治理冰河期的概念可以与其他更简单的DeFi协议中存在的 “治理封锁 “的概念进行大致的比较。但重点不是技术封锁–由于其复杂性，在Maker中是不可能的–而是政治和官僚的封锁。

拥有基于规范和规则的治理冰河时代是非常有益的，这些规范和规则是精心制定的，并且仍然与外部现实相关，因为它带来了经济和商业环境需要的稳定性，以吸引大规模的发展和投资。由于在足够长的时间尺度上，治理的冰河时代最终总是应该失去相关性，所以自然会有长期的冰河时代，直到外部现实发生足够的变化，由短期的治理地震打破，以适应新的现实并触发新的冰河时代的条件。

在造物主中，这种动态还没有得到很好的发展，自从基金会开始溶解以来，我们就一直处于一个拉长的地震。然而我不认为现在还没有一条明确的通往冰河时代的道路。最初的MIP框架试图通过MIP1在某种程度上规划如何实现冰河时代，它阐明了MIP应该涵盖的各个领域，然后一旦完成了这些，MIP过程将变得更加严格和缓慢，但它在试图预测创客治理需要什么功能时为时过早，被遗忘了。

今天Maker治理中最大的问题是核心单位的统治和缺乏透明度标准，这助长了更广泛和不太了解MKR持有人社区的普遍看法（例如你会在reddit、twitter或区域社交媒体上发现的），即核心单位效率低下，甚至腐败。这当然是错误的，绝大多数在核心单位工作的人都是这个领域的顶级专业人士，他们已经提供了大量的价值。

然而，如果MKR持有人没有看到价值，并感到沮丧和被排除在过程之外，这些都不重要。

尽管存在不足和负面看法，但核心单位能够发展到今天的规模，接管并很好地运行协议，为创客治理官僚机构的下一步发展提供了无比强大的基础，这真是一个奇迹。

现在是时候采取下一个合乎逻辑的步骤了，通过刻意设计一个治理冰期：人马座时代，将重心移回MKR持有者手中，在未来50年MakerDAO努力应对气候紧急情况并进行超速增长时，它可以作为一颗北极星。

我建议治理冰河时代以代币经济学引擎命名，该引擎将在其持续时间内被MKR发行所推动，因为我相信代币经济学是DAO中最强大的力量。通过明确调整tokenomics和治理机制，我们拥有稳定平衡的最佳条件。

打造政治和官僚的平衡点

人马座时代从根本上意味着将内部结构落实到位，以支持清洁货币的外部愿景，其官僚机构和政治动态有足够的弹性来克服所有挑战，并可信地产生一个长期稳定的环境，同时仍有必要的灵活性来适应未来几十年发生的变化或不可预见的事件。

我建议从治理政治的角度出发，通过让MKR持有者和代表自我组织起来，在一个分散的、透明的和政治上强大的结构中创建团体，我称之为分散的选民委员会（DVC），来精心设计这个新的现实。DVC将是无许可的，因为任何人都可以创建它们，对于什么算作和不算 “正式的DVC “没有严格的定义，但一般都遵循准则，以确保它们最大限度地发挥潜力，产生影响并代表MKR持有人的利益。

• 在任何人都可以加入和观察的公开电话中，保持所有关键的沟通和决策完全透明

• 证明他们所拥有的投票权的数量，使之有可能评估一个委员会应该如何认真对待，以及他们在多大程度上可以实际影响投票结果

• 通过一个内部的、独立的决策过程进行自我管理

• 专注于他们能够合理分析和理解的接近核心单位的特定专业知识或兴趣领域

• 在治理过程中，除了他们作为MKR持有人的利益之外，不接受任何补偿，也没有直接的特殊特权。人马座引擎应该有助于弥合这一差距，为那些长期锁定的MKR持有者提供强有力的激励，使其积极参与治理，以保护其锁定的MKR。

• 作为生态系统中的核心单位和私人行为者的对手，采取权力分离的方法，认识到这些团体虽然为DAO工作或与DAO合作，但最终有与MKR持有人不同的利益和激励，这需要保持控制。

• 排除授权行为人，或通过核心单位预算获得资金的私人行为人，或通过治理权限（如抵押品上柜）赚钱的私人行为人，以保持分散的投票委员会专注于考虑MKR持有人的利益。

• 在正式场合与授权行为人和私人行为人互动，邀请他们参加电话会议，以便让他们分享信息、提供意见或建议，同时注意分权，做到完全透明。

• 监测并公布有关创客治理的政治和权力动态的信息

• 监测主持人的表现、记录和保留或流失的可能性，并考虑潜在的替代者

• 提出治理建议和MIPs

• 产生对现有MIP的意见

• 产生关于主持人建议和核心单位预算建议的意见

• 分析并公布关于核心单位和治理过程中不同层面的积极和消极成就的信息

有了分散的选民委员会，然后就有可能开始工作，建立能够创造射手座时代的基本框架。我计划个人直接参与DVCs至少一年，这是我认为完成治理地震阶段所需的时间，然后当我们开始进入治理冰河时代时，我将在帮助孵化DVCs后退出对创客治理的深入参与，这些DVCs随后可以成为保持对目标的关注的力量，确保官僚机构的不同层次在冰河时代发挥应有的作用，如果它们不这样做，最终将采取行动。

最高层次的是世界地图。世界地图是一个概念，它将所有的流程、规则、规范、要求、框架等列在一个地方，涵盖了Maker Governance所需要的一切，从头到尾，在射手座冰河时代的范围内运作和操作，以实现清洁货币的愿景，为未来50年提供一个稳定的框架。

另一个必须做出的关键决定是建立一个有界限的未来重大技术创新的范围，这些创新将被添加到制造者协议中，以便技术工作可以开始主要集中在升级和部署现有的创新，而不是发明新的创新。

设计一个 “由MKR持有者，为MKR持有者 “的创客治理，最重要的方面之一是让核心单位的官僚主义得到控制，对那些拥有创客治理所赋予的重大信任和权力的人实行非常高的透明度。

这可以通过全面的框架来实现，如权力分离框架，以确保治理过程的层级分散，以及决策影响评估框架，该框架可以规范关键绩效指标和优先级进程所需的详细程度，以及授权行为人的相关信息披露。

上述两个框架都是我认为社区需要公开讨论和采用的概念。

要做到这一点，同时又不破坏目前正在进行的所有良好工作，就需要增加核心单位的总体预算，并创建新型的治理基础设施，包括授权代表以有竞争力的工资和预算实现专业化，并创建 “元核心单位”，其作用是对其他核心单位进行专业支持和分析，以帮助弥补治理和实地工作之间的差距，例如，向DVC报告和提供数据，并在论坛或社交媒体上直接向社区报告。所有这些都将需要比现在使用的更多的资源，但它也将导致更高的整体效率，从而产生更多的净效应。

一个能够激励社区成员和市场参与者参与项目的强大愿景，以及能够帮助将其转化为可靠的现金流的强大的tokenomics，将使其有可能在短期内增加预算以完成循环，然后提供必要的资源来发展治理基础设施并创造指数级的增长和回报，在现实世界中以具体成果实现愿景。这将是一个自我实现的预言，如果社区相信这种雄心勃勃的东西实际上是我们今天所拥有的一个现实的下一站，并能围绕采取果断行动进行协调。

最终，只有当整个社区的MKR持有者聚集在一起，讨论和决定我们今天所拥有的知识的最佳解决方案，然后致力于它，才有可能创造一个可靠和稳定的冰河时代，为意识驱动的超增长提供条件。