GitHub - poetiq-ai/poetiq-arc-agi-solver: This repository allows reproduction of Poetiq’s record-breaking submission to the ARC-AGI-1 and ARC-AGI-2 benchmarks. https://share.google/8g7TP1bFiz3rz9cki

https://poetiq.ai/posts/arcagi_announcement/

https://x.com/i/status/1997366836015366577

So code is the final tool to generalize, to abstract and the machine that solves intelligence problem. And LLM can help write the code, to evaluate the results, to evolve the code, till problem solved.

So basically we have a program that calls LLM, to generate code to solve the problem.

It has a scorer. It has results error feedback, it has some initial sample code.

For next step sampling, it basically add prev successful code as new examples to the prompt.

So it’s like a RL algorithm, but in test time.

So likely arc agi test will soon be solved.

<
Previous Post
Further convergence to code
>
Next Post
ftc driver station app