from Hacker News

Teaching LLMs to Refuse Unknown Questions (TLDR: Train on "I Dunno" as Well)

by wizzard0 on 11/19/23, 8:24 PM with 1 comments

  • by wizzard0 on 11/19/23, 8:24 PM

    > Our research is motivated by the observation that previous instruction tuning methods force the model to complete a sentence no matter whether the model knows the knowledge or not.

    And so... adding the random samples that include "I don't know" to the training datasets brings the accuracy on unknown questions from <10% to over 90%