CURIO adds curiosity-driven test-time learning to LLMs, up to 18.3% gains on math discovery
UIUC-CS · hf · 2026-10-14
UIUC, University of Maryland and Stanford researchers propose CURIO, a curiosity-driven test-time learning framework that lets LLMs update themselves from successes and failures during open-ended discovery, beyond what frozen-model search allows.
- Introduces an Intrinsic Curiosity World Model (ICWM) that learns transitions in the policy's hidden-state representation and supplies prediction-error bonuses at tokens outside the policy's top-k choices
- Epoch normalization and an annealed weight regulate the bonuses' contribution, avoiding premature suppression of low-reward but promising directions
On Qwen3 backbones from 8B to 235B, CURIO beats a matched task-only RL control on five of six mathematical discovery tasks (+18.3% relative on Hadamard, +10.8% on denoising Score), matches the best reported Circle Packing result, and improves single-cell denoising at every scale, with measurably greater code diversity.
More from Research
- RFD4-Proteina's 'underrated' docs detail partial diffusion for protein redesign — AllThingsApx · 2026-10-23
- MPI-Sec PhD Thesis on Vulnerability Detection Benchmarks Includes Pre-LLM Classic "Top Score on the Wrong Exam" — mboehme_ · 2026-10-23
- LiveNerf's 29-day data finds no nerfing of Claude Opus 5.5 — TheOnlyVibemaster · 2026-10-23
- New paper makes 9 falsifiable predictions toward a major AI-enabled terror attack by 2027-2028 — StephenLCasper · 2026-10-23
- HumanoidPose recovers humanoid robots' full-body joint states from RGB video alone — cholz · 2026-10-23
- Sophont's Holon-S1 predicts genes and biomarkers from routine pathology slides — iScienceLuvr · 2026-10-23