CertJudge Reuses Evaluations With One Calibration
sanmikoyejo · x · 2026-07-12
The overview notes that LLM judges are currently used to grade AI-generated Lean 4 theorems, but the problem is "who judges the judge." Every time a judge, prompt, or threshold is changed, a new manual study is typically required. The goal of CertJudge is to use a single small-scale calibration that can be reused for all subsequent judge evaluations, achieving a correlation of roughly ρ=0.833.
Related event: CertJudge Targets Reusable Calibration for LLM Judges(3 posts)→
More from Research
- Meta FAIR's Self-Direction turns agent failures into reusable LoRA skills, +13.8pp on Terminal-Bench — karen_ullrich · 2026-10-27
- DONIC hits 86.79% on bimanual dexterous tracking, crushing prior SOTA's 9.43% — liuziwei7 · 2026-10-27
- Why don't more LLMs ship 10M-token context? A researcher asks about attention tradeoffs — agarwl_ · 2026-10-27
- PSNR below 15 dB in papers likely means the metric was misused, researcher warns — kwangmoo_yi · 2026-10-27
- Ortet's first health eval: 34 open-weight models, 50 tasks, generic rankings fall short — thegautamkamath · 2026-10-27
- Scaling-fit discrepancy: output-head FLOPs are ~65% of compute at the smallest rung, ~3% at 535B — dlwh · 2026-10-27