Looping Layers Boosts Small Decision Models: 1.4B Nears 3x Larger Model
rohanpaul_ai · x · 2026-10-14
The paper SanSi: A Looped Typed Decision Model for System 1.5 Thinking shows that for typed decision models (which output option probabilities in one pass), recursively re-applying the same layers several times before readout—"looping"—dramatically improves accuracy, which the authors call System 1.5 thinking.
- SanSi builds on Ouro-1.4B, a pretrained looped model; option probabilities are read after every loop, and every loop is trained with a proper scoring rule, so one model serves any budget from 1 to 8 loops.
- On 10,027 test decisions it reaches 72.0% accuracy, 13.5 points above a same-shape non-looped model. On liar chains deeper than anything in training it scores 72.6%, while a 4.2B single-pass model sits at chance (50%).
- Used as an RL judge without gold answers, it raises the generator's F1 by 7.7 points.
Practical takeaway: if memory limits your decision-style calls, try a looped small model before jumping to one 3x larger.
Related event: SanSi: Looped Layers Boost Small Decision Models(2 posts)→
More from Research
- Surge launches sudo L7, a benchmark measuring staff-level engineering judgment in real codebases — echen · 2026-10-14
- Eterna community event covers RNA inverse design and 3D aptamer-ligand prediction — chaitjo · 2026-10-14
- GLOW: hybrid GNN + LLM question answering over incomplete knowledge graphs — burkov · 2026-10-14
- Google's AMIE hits The Lancet, matching doctors' final diagnoses 90% of the time — SumitGup · 2026-10-14
- MIT research shows neural networks can self-organize their own learning machinery — nordicinst · 2026-10-14
- Economist compiles 100+ free social science datasets on GitHub — _akpiper · 2026-10-14