Model Releases
Jeremy's excellent work here is a great illustration of a very powerful type of approach: LLM-guided on-the-fly synthesis of a symbolic worl…
Jeremy's excellent work here is a great illustration of a very powerful type of approach: LLM-guided on-the-fly synthesis of a symbolic world model, i.e. making sense of the world by writing executabl
Jeremy's excellent work here is a great illustration of a very powerful type of approach: LLM-guided on-the-fly synthesis of a symbolic world model, i.e. making sense of the world by writing executable code that encodes your understanding of the causal mechanics of the world. So far, all of the top-performing harnesses on ARC-AGI-3 use this style of approach. Which is also the approach we recommended when we initially released the benchmark (Jeremy would know this better than most, as a former Ndea member of technical staff). I'm happy that ARC 3 has incentivized more research and more progress in this area. In one pass Opus wrote 269 programs (~12,700 lines). It built parsers for all 25 games, searching functions for 23, and game simulators for 9. It built a different harness for each problem. The filesystem memory idea came from the excellent PRO-LONG harness: https://github.com/al…
Related
- neurosymbolic world models – exactly what I argued for in 2020 in the Next Decade in AI – for the win.
- PRISM: PRior-guided Imagination Sampling in world Models
- Feedback World Model Enables Precise Guidance of Diffusion Policy
- Thinking Ahead: Foresight Intelligence in MLLMs and World Model
Source: Francois Chollet (X) | 2026-08-14