Statistician asks what LLMs really mean when they express uncertainty
sineadwilliamso · x · 2026-10-06
A statistician (@sineadwilliamso) discusses using an LLM as a component in a statistical model. She argues RL-calibrated LLMs should be better calibrated, closer to a traditional classifier trained under a proper loss — but her real interest is understanding what LLMs actually "mean" when they express uncertainty, not just achieving best calibration.
Related event: Statisticians debate what LLM uncertainty expressions really mean(2 posts)→
More from Models
- GPT reverse-engineers a philosophy of life from someone's 10/10 book list — mimi10v3 · 2026-10-06
- Eric Jang: a tier 2 release beats none, but lab must catch up fast vs DSv4 Flash — ericjang11 · 2026-10-06
- Daniel Han publishes summary of LLM benchmarks you can actually trust — danielhanchen · 2026-10-06
- Claim Verification Benchmarks Mostly Test Retrieval, Not Reasoning, Finds 24K-Trace Study — deliprao · 2026-10-06
- COLM26 study: LLMs ace claim verification benchmarks by taking shortcuts, not verifying — deliprao · 2026-10-06
- Opus 5.5 uses 26k tokens vs Astra's 12k yet costs 23% less per task at equal AA score — ChrisGPT · 2026-10-06