Minsky

Inspiration

Marvin Minsky, one of the "fathers of AI", had another big idea: there is no "self", instead the human brain is a system of hundreds of interacting neural agents.

This will likely turn out to be true model of how the brain works.

Let's use AI to identify the Neural Agents in someone's brain and generate a JSON output that they can use in visualization and simulation tools.

What it does

Feeds your uploaded data (such as your blog) into an LLM and scans for the strengths of hundreds of "neural agents". Outputs a JSON model of those agents in a person.

How we built it

• Used ChatGPT to build a large matrix with correlations of hundreds of potential neural agents
• Developed a prompt to pass in a subject's data and make API calls to claude and/or davinci for completion of a detection score from 1 to 1,000 of each neural agent
• Correlate the data with a small nodejs app
• Use Scroll, Javascript, and Simoji as demos of visualizing a brain json file

Challenges we ran into

• Some minor frustrations in using the LLM APIs (lots of trial and error and some basic bad error messages and flakiness in the APIs).
• Sending a lot of data to LLMs so context size limits, speed and cost are an issue
• I did not have time to launch a fully self-serve version but could have one out in just a few more hours.

Accomplishments that we're proud of

I wasn't sure if this would be viable. AFAIK no one has done it before, using Minsky's model of the mind.

I was surprised that it was possible to create a proof of concept in under 12 hours. LLMs have certainly unlocked a lot of possibilities for novel products.

What we learned

It is now possible to test Minsky's theories at scale in a way that wasn't possible before LLMs.

What's next for Minsky

• Do a design pass
• Test v0 with friends
• Add tests
• Add a paid version that outputs a larger model faster.
• Launch on Reddit, HN, Twitter, FB

Built With

• javascript
• multiagent
• node.js
• openai
• simulation