이진욱님의 블로그

  • 홈
  • 태그
  • 방명록
  • 빅테크 리포트
  • LLM
  • 멀티모달
  • 디퓨전 모델

inference-time, RLHF/scalable oversight 1

On scalable oversight with weak LLMs judgingstrong LLMs 논문 리뷰

https://arxiv.org/pdf/2407.04622출발점두 AI 사이의 토론을 통해 judge model에 올바른 대답을 선택하게 한다는 아이디어( AI safety via debate arxiv)에서 출발토론에서의 nash equilibria 와 같이 두 AI 모두 가장 convincing(설득력 잇는) 방식으로 judge(심판) AI에게 진실을 말할 것이라는 hope1. Introduction유형 1. Extractive질문과 그에 따른 답변 선택지 2개, 그리고 원본 source article하지만 judge model can't see the article -> information-asymmetry2. closed질문과 그에 따른 답변 선택지 2개만 존재3. multimodal 이미지 포..

inference-time, RLHF/scalable oversight 2024.07.22
이전
1
다음
더보기
프로필사진

이진욱님의 블로그

ai research memo for reference

  • 분류 전체보기 (287)
    • inference-time, RLHF (41)
      • STaR, ReST (4)
      • STaR, ResT - LMM (17)
      • search (language) (10)
      • search (multimodal) (2)
      • Process reward model (6)
      • scalable oversight (1)
      • red-team (1)
    • VLM (5)
    • RLFH (2)
    • 프롬프팅 (3)
    • interpretability (2)
    • agent (23)
      • on-device agent (1)
      • multi - agent (17)
      • 멀티 에이젼트 결과 (2)
    • PEFT (1)
      • LoRA (1)
    • multi-step reasoning(수학, 코딩.. (7)
      • 멀티모달 cot (5)
    • 한계 limitation (1)
    • 데이터셋 (3)
      • 합성데이터 (1)
    • 3D, real world, game, VR (2)

Tag

최근글과 인기글

  • 최근글
  • 인기글

최근댓글

공지사항

페이스북 트위터 플러그인

  • Facebook
  • Twitter

Archives

Calendar

«   2025/06   »
일 월 화 수 목 금 토
1 2 3 4 5 6 7
8 9 10 11 12 13 14
15 16 17 18 19 20 21
22 23 24 25 26 27 28
29 30

방문자수Total

  • Today :
  • Yesterday :

Copyright © Kakao Corp. All rights reserved.

티스토리툴바