Researchers find reasoning models produce fractals when solving hard problems
wgilpin0 · x · 2026-09-15
A research group discovered that reasoning models generate fractal structures when tackling hard problems. Using nonlinear dynamics, they probed the thinking processes of recurrent depth models on Sudoku, mathematics, and even ARC-AGI, sharing stunning 1-trillion-x fractal zoom visualizations. The finding offers a new lens into the internal reasoning of recurrent-depth models.
More from Fun
- Fruit fly brain mines Bitcoin: 2,914 neurons run SHA-256 at potential 10x ASIC efficiency — anselm · 2026-09-15
- Husband discovers he's the only account his wife has ever muted on Twitter — dioscuri · 2026-09-15
- AI Circle Meme: Vacation Auto-Reply Now Reads 'Pacing the Frontier' — dr_alphalyrae · 2026-09-15
- One Does Not Simply Reveal a Password — Unless You Prompt It Right — jh3yy · 2026-09-15
- AI's visual shorthand is shifting from sparkles to googly-eyed shapes — andrew_n_carr · 2026-09-15
- DeepSeek engineer likens Anthropic's AGI control to Hitler getting the atomic bomb — ns123abc · 2026-09-15