Nature Paper Explains the Mechanism of Large Language Model Hallucinations and Proposes an Open-Ended Scoring Rubric

Written by

in

A Georgia Tech and OpenAI team published a study in Nature explaining why large language models tend to guess when answering questions they are uncertain about. Post-training evaluations typically penalize models for answering “I don’t know” as heavily as they penalize incorrect answers, whereas an open-ended scoring rubric can provide a clear mechanism for scoring responses, thereby mitigating hallucinations.

Source:Newswise · Read the original article

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *