Search papers, labs, and topics across Lattice.
2
0
3
1
LLMs often commit to flawed hypotheses based on self-selected evidence, highlighting critical gaps in their abductive reasoning capabilities.
LLM judges can drastically change their evaluations鈥攂y up to 85%鈥攚hen provided with reference answers, revealing a critical flaw in no-reference assessments.
