====== Hallucination and Factivity ======

===== Overviews =====
  * **In Generation**
    * [[https://arxiv.org/pdf/2202.03629.pdf|Ji et al 2022 - Survey of Hallucination in Natural Language Generation]]
  * **In Large Language Models**
    * **[[https://arxiv.org/pdf/2309.01219.pdf|Zhang et al 2023 - Siren's Song in the AI Ocean: A Survey on Hallucination in Large Language Models]]**
    * [[https://arxiv.org/pdf/2309.05922.pdf|Rawte et al 2023 - A Survey of Hallucination in Large Foundation Models]]
    * [[https://arxiv.org/pdf/2309.06794.pdf|Ye et al 2023 - Cognitive Mirage: A Review of Hallucinations in Large Language Models]]
    * [[https://arxiv.org/pdf/2309.16459.pdf|Andriopoulos & Pouwelse 2023 - Augmenting LLMs with Knowledge: A Survey on Hallucination Prevention]]

===== Hallucination and Factivity in LLMs =====
  * **[[https://aclanthology.org/2022.acl-long.229.pdf|Lin et al 2022 - TruthfulQA: Measuring How Models Mimic Human Falsehoods]]**
  * [[https://arxiv.org/pdf/2206.04624.pdf|Lee et al 2022 - Factuality Enhanced Language Models for Open-Ended Text Generation]] - Prepends a topic prefix to sentences in the factual documents to make each sentence serve as a standalone fact during pretraining.
  * [[https://arxiv.org/pdf/2210.09150.pdf|Prompting GPT-3 To Be Reliable]]
  * [[https://arxiv.org/pdf/2305.13534|Zhang et al 2023 - How Language Model Hallucinations Can Snowball]]
  * **[[https://arxiv.org/pdf/2305.14251.pdf|Min et al 2023 - FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation]]**
  * [[https://arxiv.org/pdf/2309.05217.pdf|Du et al 2023 - Quantifying and Attributing the Hallucination of Large Language Models via Association Analysis]]
  * [[https://arxiv.org/pdf/2309.11495.pdf|Dhuliawala et al 2023 - Chain-of-Verification Reduces Hallucination in Large Language Models]]
  * [[https://arxiv.org/pdf/2310.00259.pdf|Zouying et al 2023 - AutoHall: Automated Hallucination Dataset Generation for Large Language Models]]
  * [[https://arxiv.org/pdf/2310.00741.pdf|Chen et al 2023 - FELM: Benchmarking Factuality Evaluation of Large Language Models]] [[https://github.com/hkust-nlp/felm|github]]
  * **[[https://arxiv.org/pdf/2311.08401.pdf|Tian et al 2023 - Fine-tuning Language Models for Factuality]]** Uses DPO to fine-tune an LLM to produce factual outputs
  * [[https://arxiv.org/pdf/2404.05904|Hong et al 2024 - The Hallucinations Leaderboard – An Open Effort to Measure Hallucinations in Large Language Models]]
  * **[[https://arxiv.org/pdf/2405.05904|Gekhman et al 2024 - Does Fine-Tuning LLMs on New Knowledge Encourage Hallucinations?]]**

Prompt to break down sentences into independent facts:
{{media:facts-prompt-arxiv-2305.14251.png}}
(from [[https://arxiv.org/pdf/2305.14251.pdf|Min 2023]])

==== Datasets ====
  * TruthQA: [[https://aclanthology.org/2022.acl-long.229.pdf|paper]] [[https://github.com/sylinrl/TruthfulQA|github]]
  * FactScore: [[https://arxiv.org/pdf/2305.14251.pdf|paper]] [[https://github.com/shmsw25/FActScore|github]]
  * FELM: [[https://arxiv.org/pdf/2310.00741.pdf|paper]] [[https://github.com/hkust-nlp/felm|github]]
  * LongFact: [[https://arxiv.org/pdf/2403.18802|paper]]

===== Related Pages =====
  * [[Automatic Fact Checking]]
  * [[modality#epistemic_modality|Factivity]]
  * [[Language Model]]
  * [[ml:Trustworthy AI]]