Unfamiliar Finetuning Examples Control How Language Models Hallucinate

• URL: http://arxiv.org/abs/2403.05612v2
• Date: Tue, 28 May 2024 23:56:14 GMT
• Title: Unfamiliar Finetuning Examples Control How Language Models Hallucinate
• Authors: Katie Kang, Eric Wallace, Claire Tomlin, Aviral Kumar, Sergey Levine,
• Abstract summary: Large language models are known to hallucinate when faced with unfamiliar queries. We find that unfamiliar examples in the models' finetuning data are crucial in shaping these errors. Our work further investigates RL finetuning strategies for improving the factuality of long-form model generations.
• Score: 75.03210107477157
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Large language models are known to hallucinate when faced with unfamiliar queries, but the underlying mechanism that govern how models hallucinate are not yet fully understood. In this work, we find that unfamiliar examples in the models' finetuning data -- those that introduce concepts beyond the base model's scope of knowledge -- are crucial in shaping these errors. In particular, we find that an LLM's hallucinated predictions tend to mirror the responses associated with its unfamiliar finetuning examples. This suggests that by modifying how unfamiliar finetuning examples are supervised, we can influence a model's responses to unfamiliar queries (e.g., say ``I don't know''). We empirically validate this observation in a series of controlled experiments involving SFT, RL, and reward model finetuning on TriviaQA and MMLU. Our work further investigates RL finetuning strategies for improving the factuality of long-form model generations. We find that, while hallucinations from the reward model can significantly undermine the effectiveness of RL factuality finetuning, strategically controlling how reward models hallucinate can minimize these negative effects. Leveraging our previous observations on controlling hallucinations, we propose an approach for learning more reliable reward models, and show that they improve the efficacy of RL factuality finetuning in long-form biography and book/movie plot generation tasks.

Related papers

• Exploring the Knowledge Mismatch Hypothesis: Hallucination Propensity in Small Models Fine-tuned on Data from Larger Models [0.1227734309612871]
  Fine-tuning language models with data from larger models can appear similar, but they hallucinate more often than larger models. One hypothesis is that fine-tuning a model on data produced by a larger model leads to a knowledge mismatch which contributes to hallucination. We show that on an unseen test set, a smaller model fine-tuned on data generated from a larger model produced more wrong answers when compared to models fine-tuned on data created by the small model.
  arXiv  Detail & Related papers  (2024-10-31T13:01:46Z)
• A Debate-Driven Experiment on LLM Hallucinations and Accuracy [7.821303946741665]
  This study investigates the phenomenon of hallucination in large language models (LLMs) Multiple instances of GPT-4o-Mini models engage in a debate-like interaction prompted with questions from the TruthfulQA dataset. One model is deliberately instructed to generate plausible but false answers while the other models are asked to respond truthfully.
  arXiv  Detail & Related papers  (2024-10-25T11:41:27Z)
• Knowledge Overshadowing Causes Amalgamated Hallucination in Large Language Models [65.32990889402927]
  We coin this phenomenon as knowledge overshadowing'' We show that the hallucination rate grows with both the imbalance ratio and the length of dominant condition description. We propose to utilize overshadowing conditions as a signal to catch hallucination before it is produced.
  arXiv  Detail & Related papers  (2024-07-10T20:37:42Z)
• Prescribing the Right Remedy: Mitigating Hallucinations in Large Vision-Language Models via Targeted Instruction Tuning [15.156359255401812]
  We propose a targeted instruction data generation framework named DFTG that tailored to the hallucination specificity of different models. The experimental results on hallucination benchmarks demonstrate that the targeted instruction data generated by our method are more effective in mitigating hallucinations compared to previous datasets.
  arXiv  Detail & Related papers  (2024-04-16T07:14:32Z)
• On Large Language Models' Hallucination with Regard to Known Facts [74.96789694959894]
  Large language models are successful in answering factoid questions but are also prone to hallucination. We investigate the phenomenon of LLMs possessing correct answer knowledge yet still hallucinating from the perspective of inference dynamics. Our study shed light on understanding the reasons for LLMs' hallucinations on their known facts, and more importantly, on accurately predicting when they are hallucinating.
  arXiv  Detail & Related papers  (2024-03-29T06:48:30Z)
• Less is More: Mitigating Multimodal Hallucination from an EOS Decision Perspective [55.41815486466186]
  Large Multimodal Models (LMMs) often suffer from multimodal hallucinations, wherein they create content that is not present in the visual inputs. In this paper, we explore a new angle of this issue: overly detailed training data hinders the model's ability to timely terminate generation. We find that the model assesses the completeness of the entire sequence by comparing the generated text with the image.
  arXiv  Detail & Related papers  (2024-02-22T13:33:13Z)
• Alleviating Hallucinations of Large Language Models through Induced Hallucinations [67.35512483340837]
  Large language models (LLMs) have been observed to generate responses that include inaccurate or fabricated information. We propose a simple textitInduce-then-Contrast Decoding (ICD) strategy to alleviate hallucinations.
  arXiv  Detail & Related papers  (2023-12-25T12:32:49Z)
• AutoHall: Automated Hallucination Dataset Generation for Large Language Models [56.92068213969036]
  This paper introduces a method for automatically constructing model-specific hallucination datasets based on existing fact-checking datasets called AutoHall. We also propose a zero-resource and black-box hallucination detection method based on self-contradiction.
  arXiv  Detail & Related papers  (2023-09-30T05:20:02Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.