This is our second ChatGPT tinker. Make sure to check out PixelWhimsy, an NFT collection developed (almost entirely) by AI.

Lots of DAOs bloomed out of DeFi summer, most wilted, but a few remain perennial. Those thriving today are a hotbed of collective opinions, decision-making, and execution.

We decided to use ChatGPT to cut away the thicket of governance and offer a glimpse into those DAOs where votes have the biggest, opinion-splitting impact.

spicy.vote is our aggregator for active and contentious DAO votes. For the casual degen, perhaps with an interest in the sociological aspect of governance, the breadth of DAOs was too wide to pick out where interesting and difficult decisions were being made. For the casual community member, it likely wasn’t worth the time to track proposals that they would be able to influence.

Brief

The task was relatively straightforward: populate a list of ongoing DAO votes and sort through them based on how closely-contested the current polls are.

Unlike our previous approach, we decided to take a more active role and use ChatGPT as a teacher. The two members of the collective with more novice coding skills would strengthen their knowledge base while the actual Tinkering devs were busy building.

As before, whatever code ChatGPT helps us churn out will be the final version. There will be no place for misshapen images and awkward formatting to hide, and it’s on us to drill down into the fixes.

Process

We had most of the Python packages installed from our previous project. We set about getting ChatGPT to create database models. We knew roughly that we wanted [proposal titles], [DAO names], [vote options], [vote percentages], and [time remaining].

The Snapshot API felt like a good place to start and a simpler integration than Tally. We fed ChatGPT all of the relevant queries based on our intended database schema.

Snapshot proposal data needed to be fetched at regular intervals. ChatGPT pulled together a manual [feed_updater] file for us that we could run locally.

It was time to start designing the app layout. We figured that it made sense to keep it simple and implement one column of proposals.

ChatGPT doesn’t work with visuals, so we had more influence here. We opted for a Spicy Pixel theme to keep in-line with our first AI hackathon project, PixelWhimsy. We later asked ChatGPT to add unique styling to the most fiery ongoing votes.

We needed to add filters to make sorting simpler. We gave ChatGPT the parameters and gave it feedback to fine-tune them.

Most of this process was run locally. We made the switch to Heroku and AWS using ChatGPT’s guidance on deployment and database management.

Insight

Our process here was very different from PixelWhimsy. We weren’t dealing with anything on-chain but we were still aiming to organize a potentially sizable pool of data.

Taking the time to implement our idea to track contentious DAO governance votes achieved its goal in cutting through the noise (or lack of it). Being able to easily pick out which DAOs are making important decisions gets right to the underbelly of the votes which could impact Web3 in the next 1-2 years.

At a community level, it is also interesting to see that DAO governance isn’t as gamed as we thought. Early token allocations do not always equal the ability to puppeteer the direction of a project. This is refreshing when you consider the track record of some of the previous Core DAOs in our list.

One of the most exciting outcomes was that both Tinkering members felt ChatGPT greatly steepened the coding learning curve. The benefit of having a knowledgeable code brain analyze and write code in front of you makes it far easier to grasp the underlying processes. The practice has a kind of paint-by-numbers effect, being actively guided through it means you have greater capacity to understand the intricacies.

There are still clear limitations too. Coding alongside ChatGPT can be as frustrating as it can be enlightening, it regularly misconstrues simple feedback which can leave you developing in circles with it.

Got a ChatGPT idea to share? We’re always excited to hear about potential new tinkers and always feel free to share them with us.