Residual-stream probes predict whether small math LLMs stay correct under meaning-preserving perturbations — from one forward pass, before generation. AIMO Interpretability Challenge 2026, Small Models Track.
Quick read
Latest capture 2026-10-29 04:00
Compare the starting count and capture dates, not just the percentage. How to read GitHub star growth. Missing history is not zero growth.
Archive status reflects the last catalog sync, not a maintenance assessment. Check recent releases, issue responses, and fit for your project with the repository shortlisting checklist.
0 paths
Agent instructions and tool configuration found in this repository.
No config files detected.
1 observed capture since 2026-10-29. Observed captures are shown by default.
Stars from first capture 0
Observed captures only
All tracked data
Observed snapshots
Observed snapshots
Nearest indexed repositories by embedding similarity.
White-box robustness probes for AIMO Interpretability: Small Track bundle, training-only validation transforms and reproducible controls.
Entry to the Predictive AI Evaluation Competition at NeurIPS 2026 (Stanford AIMS): Beta-Binomial shrinkage predictor, independent audit, reproducible protocol and public-data findings.
Fold-safe ensembling to the honest ceiling of Kaggle Playground S6E9, and a Darwin-Cage residual search with a power analysis. Nested OOF 0.946420; 47,679 evolved programs, 0 discoveries.
Rule-compliant Overture Maps coverage-gap estimator for 9,379 US tracts + a provenance bias discovery (Zindi x Humane Intelligence Bias Bounty)
Semantic ifs from open models, on a 3090 at home. Independent; not affiliated with Jev or TypeSafe.
Benchmark for tracking model capability after release.
