Dataset of 20,000 boolean statements, each labeled by two human annotators as common-knowledge-true, common-knowledge-false, or neither, created by the MIT Algorithmic Alignment Group for red-teaming language models.
Endorsements made here support MIT Algorithmic Alignment Group.
Dataset of 20,000 boolean statements, each labeled by two human annotators as common-knowledge-true, common-knowledge-false, or neither, created by the MIT Algorithmic Alignment Group for red-teaming language models.
Endorsements made here support MIT Algorithmic Alignment Group.
Project Details
Updated 05/18/26 · By grantmaking.aiCommonClaim is a publicly released dataset assembled by the MIT Algorithmic Alignment Group for the paper “Explore, Establish, Exploit: Red Teaming Language Models from Scratch,” consisting of 20,000 GPT-3-generated factual statements that have been labeled by two human annotators as common-knowledge-true, common-knowledge-false, or neither to support research on detecting and studying model dishonesty and failure modes.
People– no linked people
Updated 05/18/26 · By grantmaking.aiDiscussion
No comments yet. Be the first to share your thoughts.