서브메뉴
검색
Towards Agents Which Can Understand Rich Communication.
Towards Agents Which Can Understand Rich Communication.
- 자료유형
- 학위논문
- Control Number
- 0017163788
- International Standard Book Number
- 9798384448525
- Dewey Decimal Classification Number
- 384
- Main Entry-Personal Name
- Watkins, Olivia.
- Publication, Distribution, etc. (Imprint
- [S.l.] : University of California, Berkeley., 2024
- Publication, Distribution, etc. (Imprint
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- Physical Description
- 218 p.
- General Note
- Source: Dissertations Abstracts International, Volume: 86-04, Section: A.
- General Note
- Advisor: Abbeel, Pieter;Darrell, Trevor.
- Dissertation Note
- Thesis (Ph.D.)--University of California, Berkeley, 2024.
- Summary, Etc.
- 요약Today's AI systems are trained primarily on large datasets of input-output pairs. These agents may be able to condition on simple forms of communication (such as a language task description), but they're currently not capable of making use of the full spectrum of communication, verbal and non-verbal, which human teachers use to guide their students.This thesis makes progress on two challenges around teaching agents to understand rich communication. In Part 1, we develop algorithms which can efficiently ground real-time communication provided by humans, both non-verbal communication and several forms of language. We also enable agents to use language in a new way - guiding common-sense exploration.In Part 2, we address the challenge of teaching agents to understand communication by trusted sources while ignoring malicious instructions or facts provided by untrusted sources. We benchmark models' vulnerability to semantic prompt injection and jailbreak attacks, paving the way for future work addressing of these weaknesses we observed.
- Subject Added Entry-Topical Term
- Communication.
- Index Term-Uncontrolled
- Reinforcement learning
- Index Term-Uncontrolled
- Malicious instructions
- Index Term-Uncontrolled
- Language task description
- Added Entry-Corporate Name
- University of California, Berkeley Computer Science
- Host Item Entry
- Dissertations Abstracts International. 86-04A.
- Electronic Location and Access
- 로그인을 한후 보실 수 있는 자료입니다.
- Control Number
- joongbu:657153