New research release: looping helps up to 3 iterations, fewer placeholder tokens cut reasoning cost
nrehiew_ · x · 2026-11-21
nrehiew highlights findings from a helpful research release: more looping improves performance but with diminishing returns after 3 iterations, and fewer placeholder tokens reduce the number of reasoning tokens needed. He calls it super cool research and looks forward to trying it out.
More from Research
- Ex-DeepMind researcher: if the brain is 'deep in time', are frontier models an abomination patched by looping? — yacinelearning · 2026-11-21
- 'The convex hull of human math is bigger than we thought' — mathematicians' AI quip goes viral — ChenhaoTan · 2026-11-21
- Looped transformers: more looping boosts reasoning but with diminishing returns past 3 passes and fewer CoT tokens — nrehiew_ · 2026-11-21
- Team releases TML-d24-varloop-research on Hugging Face to study looping's effect on CoT monitorability — nrehiew_ · 2026-11-21
- Loop only the middle layers: design detail shows 1.51-1.91x compute-equivalent gain vs 1.24-1.36x for last layers — nrehiew_ · 2026-11-21
- New Company Agent Bench: Gemini Neon leads at 87.4%, simple API harnesses beat products — philhchen · 2026-11-21