ACUTE 프로토콜, 언어모델 활성화로 신뢰도 및 유용성 향상
The ACUTE Protocol: Operationalizing Language Model Activations for Better Calibration, Utility, and Trust
- 매체
- arXiv
- 게시
- 중요도
- 3/5
- 화제 지수
- 2
한국어 요약
언어모델의 신뢰성 확보를 위해 보정(calibration)과 정보성을 균형있게 평가하는 EURO 지표와 ACUTE 프로토콜을 제안했다. 6개 모델에서 객관식 질답, 도구 호출, 과학 문서 요약 등 3가지 작업에 걸쳐 기존 방법 대비 우수한 성능을 보였으며, 낮은 보정 오류를 유지한다.
원문 발췌 · English
As language models improve and become increasingly deployed to solve a variety of tasks, trustworthiness becomes essential. Calibration is a good proxy for trust: well-calibrated confidence estimates help inform the risk versus reward tradeoff when trusting a specific model output. Unfortunately, even as models improve, they remain poorly calibrated, often biasing towards overconfidence. Additionally, calibration can be gamed: a policy that always predicts the base rate is perfectly calibrated, but completely uninformative. To resolve this, we develop a new metric, expected utility renormalized by the oracle (EURO), that balances calibration and informativeness. We also propose a general-purpose activation-based confidence, utility, and trust estimation protocol (ACUTE) to appropriately adjudicate uncertainty. The ACUTE protocol provides flexible, sample-efficient, and compute-efficient confidence estimators for 3 tasks including multiple choice question answering, tool-calling, and scientific document summarization across 6 models from 4 model families. ACUTE outperforms strong baselines on EURO, while maintaining low calibration error. Taken together, our work shows that equipping LLMs with the ACUTE protocol can improve calibration, utility, and trustworthiness in numerous settings.