Study Questions Threshold-Based Abstention Rules for Open Language Models
A new arXiv paper critiques Kalai et al.'s proposal that language models should abstain unless their probability of being correct exceeds a stated threshold t, arguing that abstention is not mere threshold compliance.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-11-13 05:00 · arXiv cs.CL
Abstention Is Not Threshold Compliance: Stated Scoring Rules in Open Language Models