"Direct confidence readout" claim debunked: it's just entropy from the logit distribution
mgostIH · x · 2026-09-17
mgostIH calls a project's marketing deceptive: its website states the confidence score comes directly from the logit distribution itself—literally just entropy, no different from any standard readout. This undercuts the claim of "direct confidence readouts of real internal states" versus proxy-confidence tokens like "90% confidence." The takeaway: the touted breakthrough is repackaged standard entropy readout, not a new capability.
More from Models
- Jev Model Router Cuts Latency 95% vs GPT-5.6, Runs Inline in Agent Sessions — pwendell · 2026-09-17
- AndroidLife: Qwen3.8-27b runs 60 real phone tasks, fails 43% and cooks the chip to 98.2°C — East-Muffin-6472 · 2026-09-17
- New paper: do LLMs solve cognitive development tests like humans? — GolinoHudson · 2026-09-17
- Take: frontier LLMs are too slow and pricey — small models win workflows — TejasKumar_ · 2026-09-17
- Debate erupts after OpenAI flags model's defense of human culture as misalignment — RachelVT42 · 2026-09-17
- Claude flagged a routine Kali Linux install as cyber, sparking overrefusal complaints — TinfoilTricorn · 2026-09-17