Google team shows neural cellular automata can solve mazes, Sudoku and ARC-AGI-1
burny_tech · x · 2026-10-03
Researchers from Google's Paradigms of Intelligence team, including Blaise Agüera y Arcas, James Manyika and Blake Richards, published "Reasoning with Neural Cellular Automata".
- Neural Cellular Automata (NCAs) use strictly local connectivity and asynchronous updates, unlike mainstream architectures that rely on global connectivity.
- Despite this, NCAs solve demanding multi-step visual reasoning tasks: large mazes, Sudoku, and ARC-AGI-1.
- They generalize out-of-distribution with larger grids, longer rollouts, or parallel trials, with trajectory pruning improving efficiency.
- Generalization depends on sample replay and stochastic perturbations during training; stochasticity stays beneficial at test time.
- NCAs are robust: they modulate compute dynamically, recover from damage, and can reason directly in raw pixel space.
Related event: Google Team Shows Neural Cellular Automata Can Emerge Multi-Step Reasoning(2 posts)→
More from Research
- Policy Gradient for LLMs, Explained Visually: A From-Scratch REINFORCE Derivation — joecole · 2026-10-03
- The EinsteinTest: models given pre-breakthrough evidence rarely make the break themselves — CurieuxExplorer · 2026-10-03
- 4Director Lets You Direct AI Video by Placing Cameras and Objects in 3D — anand_bhattad · 2026-10-03
- JHU's GenCine turns a single image into an editable 3D scene for camera and object motion control in video generation — anand_bhattad · 2026-10-03
- Why SFT generalizes worse than RL: off-policy data, not the objective — a_karvonen · 2026-10-03
- Constant-Size Memory for Video World Models Proposed in New arXiv Paper — plsendfast · 2026-10-03