Scientia AI●

a bilingual, research-first reading log for AI

auth

← back to the log

arXiv q-bio + cs.AI: clinical/medical

A decodability criterion predicts when hidden-state selection beats majority voting in large language models

디코더빌리티 기준이 대규모 언어 모델에서 숨은 상태 선택이 다수결 투표를 이기는 시점을 예측한다

2026.08.17 · biomedical-ai

Read the original article ↗

No write-up yet — this entry is still just the auto-fetched hook. Aaron's own summary (200–500 words) will replace this note once he's written it.