Neural Cellular Automata show a visual chain-of-thought, solving mazes like slime molds
GregKamradt · x · 2026-10-02
In a 9-part thread, Mayalen Etcetera shows how local communication in Neural Cellular Automata (NCAs) produces a visual "chain of thought": on mazes, NCAs explore paths in parallel and backtrack from dead ends, much like slime molds, and on ARC-AGI-1 tasks, colors and shapes gradually move across the grid to solve problems. The work makes model reasoning directly observable as spatial dynamics.
More from Research
- Arc Institute Builds High-Quality Perturb-seq Data for the 2026 Virtual Cell Challenge — anshulkundaje · 2026-10-02
- Frontier AI models now beat CPAs on accounting tasks that humans won as of May 2025 — soumitrashukla9 · 2026-10-02
- Goodfire proposes VPD: decomposing LM parameters into faithful subcomponents — Sauers_ · 2026-10-02
- "Virtual structures": Goodfire shares new parameter interpretation work — Sauers_ · 2026-10-02
- NavVerse benchmark tests robots on indoor-to-outdoor navigation across 200 scenes and 10K episodes — GhaffariMaani · 2026-10-02
- Anima Anandkumar to keynote SC26 on AI for simulating the physical world — AnimaAnandkumar · 2026-10-02