Reasoning slowdowns are transient chaos: paper shows models have fractal basins of attraction
wgilpin0 · x · 2026-09-09
A new arXiv paper (2609.04963) by Jeffrey Lai, Anthony Bao, John Quinn and William Gilpin shows reasoning models exhibit transient chaos — a physical consequence of the computational complexity of hard tasks. Across Sudoku, mazes, visual puzzles and math logic, leading models behave as dynamical systems with fractal basins, fractality increasing with difficulty. Chaos emerges as reasoning gets trapped near saddle points corresponding to nearly-correct solutions, making slowdowns an inevitable cost of problem hardness.
Related event: Study Explains LLM Overthinking via Fractal Basins(13 posts)→
More from Models
- Model Can't Draw ASCII Whales, So It Writes a Node.js Script to Improve — teortaxesTex · 2026-09-09
- Blender head-to-head: same prompt, 12 seconds, and one frontier model is in a different class — ZeroStateReflex · 2026-09-09
- Netlify adds GPT-6 Astra, Gemini 3.8 Flash, Claude Fable 5.1 and smarter Agent Runner scoping — thisiskp_ · 2026-09-09
- GLM-5.3-Flash tops agentic tool-call leaderboard at 78%, priced at just $0.50/M output tokens — shensi · 2026-09-09
- Artificial Analysis Updated Benchmarks Twice in 4 Days for Astra — py-net · 2026-09-09
- Gary Marcus: it was the harness, not the model — hidden risk of opaque scaffolding — GaryMarcus · 2026-09-09