Certifying LLM Judges via Behavioral Checks
sanmikoyejo · x · 2026-07-12
The core idea is to stop constantly re-annotating, and instead "certify" judges using automated, falsifiable behavioral checks. The paper lists four categories of checks:\n\n1. P1 Identity\n2. P2 Bug monotonicity\n3. P3 Spec monotonicity\n4. P4 Stability\n\nBy applying controlled perturbations to arbitrary Lean files, this method verifies whether a judge is reliable without the need for gold-standard labels.
More from Research
- MCP fault injection study: agent task scores drop 60% under realistic tool failures, scale doesn't help — dair_ai · 2026-10-26
- Study: AI reviewing AI misses most reward hacking; best models miss 25%+ of hacks — ronbodkin · 2026-10-26
- FLUX.3 Klein step-distilled to 4-8 steps with CFG baked in — Fit-Structure-6489 · 2026-10-26
- NLI alone rejects 5/6 new agent memories; a 0.45 similarity gate fixes it all — Alternative_Pin9598 · 2026-10-26
- AI folks, stop saying "biology" when you mean drug development — KevinKaichuang · 2026-10-26
- ostris pre-training "Primal VAE": 1-channel latents match modern 8x VAEs via patching — ostrisai · 2026-10-26