Small LLM Judges Approved 11% and 41% of Wrong Answers. Then I Fixed My Own Pairwise Test.
This story is from 2026-10-07. It is preserved in the archive; the latest stories are on the live feed.
Grading small open judges against deterministic oracles, plus the harness that makes most judge calls unnecessary. Scope note, first paragraph as promised: every judge here is ≤3B parameters running locally. Nothing below carries over to frontier judges. Synthetic corruptions and natural errors are…
Read the full story at DEV Community — Machine Learning ↗
Timeline · 1 report
- 2026-10-07 14:32 · DEV Community — Machine Learning
Small LLM Judges Approved 11% and 41% of Wrong Answers. Then I Fixed My Own Pairwise Test.