Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
119 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Redefining "Hallucination" in LLMs: Towards a psychology-informed framework for mitigating misinformation (2402.01769v1)

Published 1 Feb 2024 in cs.CL and cs.AI

Abstract: In recent years, LLMs have become incredibly popular, with ChatGPT for example being used by over a billion users. While these models exhibit remarkable language understanding and logical prowess, a notable challenge surfaces in the form of "hallucinations." This phenomenon results in LLMs outputting misinformation in a confident manner, which can lead to devastating consequences with such a large user base. However, we question the appropriateness of the term "hallucination" in LLMs, proposing a psychological taxonomy based on cognitive biases and other psychological phenomena. Our approach offers a more fine-grained understanding of this phenomenon, allowing for targeted solutions. By leveraging insights from how humans internally resolve similar challenges, we aim to develop strategies to mitigate LLM hallucinations. This interdisciplinary approach seeks to move beyond conventional terminology, providing a nuanced understanding and actionable pathways for improvement in LLM reliability.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (42)
  1. Abid, A. Is chatgpt getting dumber? Deutsche Welle, 2023.
  2. Persistent anti-muslim bias in large language models. In Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society, pp.  298–306, 2021.
  3. Artificial hallucinations in chatgpt: implications in scientific writing. Cureus, 15(2), 2023.
  4. Hallucinations: theoretical and clinical overview. The American journal of psychiatry, 143(9):1088—1097, September 1986. ISSN 0002-953X. doi: 10.1176/ajp.143.9.1088. URL https://doi.org/10.1176/ajp.143.9.1088.
  5. The recency effect: Implicit learning with explicit retrieval? Memory & Cognition, 21:146–155, 1993.
  6. Yolov4: Optimal speed and accuracy of object detection, 2020.
  7. How is chatgpt’s behavior changing over time?, 2023.
  8. Dodig-Crnkovic, G. How gpt realizes leibniz’s dream and passes the turing test without being conscious, Aug 2023. URL https://www.mdpi.com/2813-0324/8/1/66.
  9. Edwards, B. Why chatgpt and bing chat are so good at making things up. Ars Technica. https://arstechnica. com/informationtechnology/2023/04/why-ai-chatbots-are-the-ultimate-bs-machines-and-how-people-hope-to-fix-them, 2023.
  10. Delusions, illusions and hallucinations in epilepsy: 1. elementary phenomena. Epilepsy Research, 85(2):162–171, 2009. ISSN 0920-1211. doi: https://doi.org/10.1016/j.eplepsyres.2009.03.018. URL https://www.sciencedirect.com/science/article/pii/S0920121109000813.
  11. Festinger, L. A theory of cognitive dissonance, 1957. doi: 10.1515/9781503620766.
  12. Flavell, J. H. Metacognition and cognitive monitoring: A new area of cognitive–developmental inquiry. American psychologist, 34(10):906, 1979.
  13. Herrman, J. Is chatgpt getting dumber? New York Magazine, 2023.
  14. Survey of hallucination in natural language generation, Nov 2022. URL https://arxiv.org/abs/2202.03629.
  15. Towards mitigating hallucination in large language models via self-reflection, 2023.
  16. The role of metacognitive components in creative thinking. Frontiers in psychology, 10:2404, 2019.
  17. Source monitoring. Psychological bulletin, 114(1):3, 1993.
  18. Klayman, J. Varieties of confirmation bias. Psychology of learning and motivation, 32:385–418, 1995.
  19. Lai, E. R. Metacognition: A literature review. Always learning: Pearson research report, 24:1–40, 2011.
  20. Where’s the liability in harmful ai speech? SSRN Electronic Journal, 2023. doi: 10.2139/ssrn.4531029.
  21. Towards understanding and mitigating social biases in language models. In International Conference on Machine Learning, pp.  6565–6576. PMLR, 2021.
  22. Entity-aware image caption generation, Nov 2018. URL https://arxiv.org/abs/1804.07889.
  23. Towards logiglue: A brief survey and a benchmark for analyzing logical reasoning capabilities of language models. arXiv preprint arXiv:2310.00836, 2023.
  24. Complex visual hallucinations. Clinical and neurobiological insights. Brain, 121(10):1819–1840, 10 1998. ISSN 0006-8950. doi: 10.1093/brain/121.10.1819. URL https://doi.org/10.1093/brain/121.10.1819.
  25. Biases in large language models: Origins, inventory, and discussion. J. Data and Information Quality, 15(2), jun 2023. ISSN 1936-1955. doi: 10.1145/3597307. URL https://doi.org/10.1145/3597307.
  26. NHS. Hallucinations and hearing voices. National Health Service choices, 2023. URL https://www.nhs.uk/mental-health/feelings-symptoms-behaviours/feelings-and-symptoms/hallucinations-hearing-voices/.
  27. OpenAI. Gpt-4 technical report, 2023.
  28. Schacter, D. L. The seven sins of memory: Insights from psychology and cognitive neuroscience. American psychologist, 54(3):182, 1999.
  29. Misattribution, false recognition and the sins of memory. Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences, 356(1413):1385–1393, 2001.
  30. Retrieval without recollection: An experimental analysis of source amnesia. Journal of Verbal Learning and Verbal Behavior, 23(5):593–611, 1984. ISSN 0022-5371. doi: https://doi.org/10.1016/S0022-5371(84)90373-6. URL https://www.sciencedirect.com/science/article/pii/S0022537184903736.
  31. Schott, G. D. Exploring the visual hallucinations of migraine aura: the tacit contribution of illustration. Brain, 130(6):1690–1703, 01 2007. ISSN 0006-8950. doi: 10.1093/brain/awl348. URL https://doi.org/10.1093/brain/awl348.
  32. Ease of retrieval as information: Another look at the availability heuristic. Journal of personality and social psychology, 61(2):195–202, 08 1991. URL https://utk.idm.oclc.org/login?url=https://www.proquest.com/scholarly-journals/ease-retrieval-as-information-another-look-at/docview/614382983/se-2.
  33. Hallucination or confabulation? neuroanatomy as metaphor in large language models. PLOS Digital Health, 2023.
  34. Llama 2: Open foundation and fine-tuned chat models, 2023.
  35. A stitch in time saves nine: Detecting and mitigating hallucinations of llms by validating low-confidence generation. arXiv preprint arXiv:2307.03987, 2023.
  36. Investigating gender bias in language models using causal mediation analysis. Advances in neural information processing systems, 33:12388–12401, 2020.
  37. Auditory Hallucinations in Schizophrenia and Nonschizophrenia Populations: A Review and Integrated Model of Cognitive Mechanisms. Schizophrenia Bulletin, 38(4):683–693, 03 2012. ISSN 0586-7614. doi: 10.1093/schbul/sbs045. URL https://doi.org/10.1093/schbul/sbs045.
  38. Confabulation. Confabulation - StatPearls - NCBI Bookshelf, 2023.
  39. Evaluating reading comprehension exercises generated by llms: A showcase of chatgpt in education applications, 2023. URL https://aclanthology.org/2023.bea-1.52.pdf.
  40. Siren’s song in the ai ocean: A survey on hallucination in large language models, Sep 2023. URL https://arxiv.org/abs/2309.01219.
  41. Reducing quantity hallucinations in abstractive summarization. arXiv preprint arXiv:2009.13312, 2020.
  42. Detecting hallucinated content in conditional neural sequence generation. arXiv preprint arXiv:2011.02593, 2020.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (3)
  1. Elijah Berberette (1 paper)
  2. Jack Hutchins (4 papers)
  3. Amir Sadovnik (9 papers)
Citations (6)
View on Semantic Scholar

Summary

We haven't generated a summary for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize Now
X Twitter Logo Streamline Icon: https://streamlinehq.com

Tweets