Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • NEWS AND VIEWS

‘Fighting fire with fire’ — using LLMs to combat LLM hallucinations

The number of errors produced by an LLM can be reduced by grouping its outputs into semantically similar clusters. Remarkably, this task can be performed by a second LLM, and the method’s efficacy can be evaluated by a third.
  1. Karin Verspoor

    1. Karin Verspoor is in the School of Computing Technologies, RMIT University, Melbourne, Victoria 3000, Australia and in the School of Computing and Information Systems, University of Melbourne, Melbourne, Victoria 3010, Australia.

    View author publications

    You can also search for this author in PubMed  Google Scholar

Text-generation systems powered by large language models (LLMs) have been enthusiastically embraced by busy executives and programmers alike, because they provide easy access to extensive knowledge through a natural conversational interface. Scientists too have been drawn to both using and evaluating LLMs — finding applications for them in drug discovery1, in materials design2 and in proving mathematical theorems3. A key concern for such uses relates to the problem of ‘hallucinations’, in which the LLM responds to a question (or prompt) with text that seems like a plausible answer, but is factually incorrect or irrelevant4. How often hallucinations are produced, and in what contexts, remains to be determined, but it is clear that they occur regularly and can lead to errors and even harm if undetected. In a paper in Nature, Farquhar et al.5 tackle this problem by developing a method for detecting a specific subclass of hallucinations, termed confabulations.

Access options

Rent or buy this article

Prices vary by article type

from$1.95

to$39.95

Learn more

Prices may be subject to local taxes which are calculated during checkout

Nature 630, 569-570 (2024)

doi: https://doi.org/10.1038/d41586-024-01641-0

References

  1. Vert, J.-P. Nature Biotechnol. 41, 750–751 (2023).

    Article  PubMed  Google Scholar 

  2. Jablonka, K. M. et al. Digit. Discov. 2, 1233–1250 (2023).

    Article  PubMed  Google Scholar 

  3. Frieder, S. et al. Mathematical capabilities of ChatGPT. In Proc. NeurIPS 36 (eds Oh, A. et al.) (NIPS, 2023).

    Google Scholar 

  4. Hicks, M. T., Humphries, J. & Slater, J. Ethics Inf. Technol. 26, 38 (2024).

    Article  Google Scholar 

  5. Farquhar, S., Kossen, J., Kuhn, L. & Gal, Y. Nature 630, 625–630 (2024).

    Article  Google Scholar 

  6. Firth, J. R. Studies in Linguistic Analysis (Blackwell, 1957).

    Google Scholar 

  7. Landauer, T. K. & Dumais, S. T. Psych. Rev. 104, 211–240 (1997).

    Article  Google Scholar 

  8. Bender, E. M. & Koller, A. in Proc. 58th Ann. Meet. ACL 5185–5198 (Association for Computational Linguistics, 2020).

    Google Scholar 

  9. Mitchell, M. & Krakauer, D. C. Proc. Natl Acad. Sci. USA 120, e2215907120 (2023).

    Article  PubMed  Google Scholar 

  10. Zhang, T., Kishore, V., Wu, F., Weinberger, K. Q. & Artzi, Y. In 8th Int.Conf.Learning Represent. (ICLR, 2020); available at https://openreview.net/forum?id=SkeHuCVFDr

    Google Scholar 

  11. Wang, L. L. et al. In Proc. 61st Ann. Meet. ACL Vol. 1, 9871–9889 (Association for Computational Linguistics, 2023).

    Google Scholar 

  12. Sun, T., He, J., Qiu, X. & Huang, X. In Proc. 2022 Conf. Empirical Methods in Natural Language Processing 3726–3739 (Association for Computational Linguistics, 2022).

    Google Scholar 

  13. Koike, R., Kaneko, M. & Okazaki, N. Proc. AAAI Conf. Artificial Intell. 38, 21258–21266 (AAAI, 2024).

    Article  Google Scholar 

  14. Li, Y. et al. Preprint at arXiv https://doi.org/10.48550/arXiv.2405.12689 (2024).

  15. Taloni, A., Scorcia, V. & Giannaccare, G. Eye 38, 397–400 (2024).

    Article  PubMed  Google Scholar 

  16. Zhang, Y. et al. Detection vs. Anti-detection: Is Text Generated by AI Detectable? In Wisdom, Well-Being, Win-Win (eds Sserwanga, I. et al.) Lecture Notes in Computer Science Vol. 14596 (Springer, 2024).

    Google Scholar 

Download references

Competing Interests

K.V. has received speaker fees and travel reimbursement for presentations on Artificial Intelligence, Natural Language Processing/LLMs, and AI in Health care; research funding from the Australian Research Council, the Australian National Health and Medical Research Council, and the Medical Research Futures Fund, and has research partnerships with Elsevier BV. K.V. is co-founder and Victoria Node Lead of the Australian Alliance for Artificial Intelligence in Healthcare; and a member of the Standards Australia Committee, IT-014-21, AI in Healthcare.

Subjects

Latest on:

Nature Careers

Jobs

  • Endowed Chair in Macular Degeneration Research

    Dallas, Texas (US)

    The University of Texas Southwestern Medical Center (UT Southwestern Medical Center)

  • Postdoctoral Fellow

    Postdoc positions on ERC projects – cellular stress responses, proteostasis and autophagy

    Frankfurt am Main, Hessen (DE)

    Goethe University (GU) Frankfurt am Main - Institute of Molecular Systems Medicine

  • ZJU 100 Young Professor

    Promising young scholars who can independently establish and develop a research direction.

    Hangzhou, Zhejiang, China

    Zhejiang University

  • Qiushi Chair Professor

    Distinguished scholars with notable achievements and extensive international influence.

    Hangzhou, Zhejiang, China

    Zhejiang University

  • Research Postdoctoral Fellow - MD

    Houston, Texas (US)

    Baylor College of Medicine (BCM)

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing

Search

Advanced search

Quick links