Aligning Evaluation Incentives to Reduce Hallucination @SimonsInstitute
Aligning Evaluation Incentives to Reduce Hallucination  @SimonsInstitute
Uploaded October 2026 | Updated November 2026, 2 days ago
Adam Kalai ( RESI)
https://simons.berkeley.edu/talks/adam-kalai-resi-2026-10-07
Trustworthy AI: From Hallucinations to Reliable Autonomy

Large language models are typically evaluated using accuracy-based metrics that reward confident guessing while giving little or no credit for acknowledging uncertainty. Because evaluations influence which models and methods are selected, these incentives can work against techniques that reduce hallucinations at the cost of answering fewer questions. The talk discusses approaches to evaluation that better reward useful uncertainty and reflect the differing costs of mistakes, correct answers, and abstentions, along with open questions in designing evaluations whose incentives better match real-world objectives.

Based on joint work with Santosh Vempala, Ofir Nachum, and Edwin Zhang.
Aligning Evaluation Incentives to Reduce HallucinationATHENA: Steering Generative Models toward Better Visual ReasoningReliable Language GenerationStrong Convergence (Part I)Expanding GroupsLinear hashing and friendsMy Favorite Expanders | Richard M. Karp Distinguished LectureMatching Correlated Random Geometric GraphsFault-Tolerant Quantum Computation with Adversarial ErrorsApollonian random manifolds and their bass notesSampling from the Sherrington-Kirkpatrick model up to β<1/2Coboundary Expansion and Locally Testable Codes (Part 2)
Simons Institute for the Theory of Computing |

Aligning Evaluation Incentives to Reduce Hallucination

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER