LLM reprograms neural cellular automata by editing weights, no gradient training
burny_tech · x · 2026-11-04
A research report from Stefano Nichele (Østfold University / Simula AI Safety) shows GPT-6 (Astra) directly editing the 8,336 shared weights of a growing-gecko neural cellular automaton. Six zero-feedback edits achieved numerically precise color/spatial symmetries and doubled maturation time; a feedback stage used 36 adaptive evaluations to build a bounded timer inside the network. The edited automaton stayed stable across 8 unseen 8,192-step runs, demonstrating reproducible weight-only reprogramming without gradient-based training.
More from Research
- NRGPT paper reimagines GPT inference as gradient descent on an explicit energy landscape — burkov · 2026-11-04
- Fireworks event slides show sliding-window bit-sharing quantization trick — sophiamyang · 2026-11-04
- UCLA Study Seeks Users of 'Griefbots' to Interview on AI and Mourning ($30 Gift Card) — Interesting-Cat-7838 · 2026-11-04
- Tiny LoRA adapter beats DNA model Carbon-X on RareBench, exposing generalization limits — danielmckinn0n · 2026-11-04
- QuickProof: formal verification for C with specs written as plain C tests — thomasahle · 2026-11-04
- RSNA 2026 Deep Learning Lab to teach medical imaging VLM dataset building from scratch — PRouzrokh · 2026-11-04