English

Does Fine-Tuning LLMs on New Knowledge Encourage Hallucinations?

Computation and Language 2024-10-02 v3

Abstract

When large language models are aligned via supervised fine-tuning, they may encounter new factual information that was not acquired through pre-training. It is often conjectured that this can teach the model the behavior of hallucinating factually incorrect responses, as the model is trained to generate facts that are not grounded in its pre-existing knowledge. In this work, we study the impact of such exposure to new knowledge on the capability of the fine-tuned model to utilize its pre-existing knowledge. To this end, we design a controlled setup, focused on closed-book QA, where we vary the proportion of the fine-tuning examples that introduce new knowledge. We demonstrate that large language models struggle to acquire new factual knowledge through fine-tuning, as fine-tuning examples that introduce new knowledge are learned significantly slower than those consistent with the model's knowledge. However, we also find that as the examples with new knowledge are eventually learned, they linearly increase the model's tendency to hallucinate. Taken together, our results highlight the risk in introducing new factual knowledge through fine-tuning, and support the view that large language models mostly acquire factual knowledge through pre-training, whereas fine-tuning teaches them to use it more efficiently.

Keywords

Cite

@article{arxiv.2405.05904,
  title  = {Does Fine-Tuning LLMs on New Knowledge Encourage Hallucinations?},
  author = {Zorik Gekhman and Gal Yona and Roee Aharoni and Matan Eyal and Amir Feder and Roi Reichart and Jonathan Herzig},
  journal= {arXiv preprint arXiv:2405.05904},
  year   = {2024}
}

Comments

Accepted as a long paper at EMNLP 2024

R2 v1 2026-06-28T16:22:21.917Z