AI hallucinations can’t be stopped — but these techniques can limit their damage

2025/01/21 by Nicola Jones, N.B. Jones · 1 voice · 2 citations
Neuroscience · #Neuroethics, Human Enhancement, Biomedical Innovations

paper · doi:10.1038/d41586-025-00068-5

crossref issued 2025/01/21 · crossref published 2025/01/21 · crossref published-online 2025/01/21 · openalex publication_date 2025/01/21 · crossref created 2025/01/21 · crossref deposited 2025/01/21 · crossref published-print 2025/01/23 · openalex created_date 2025/10/10 · crossref indexed 2026/07/30 · openalex updated_date 2026/07/31

Citations

• Representation Engineering: A Top-Down Approach to AI Transparency
• Calibrated Language Models Must Hallucinate
• Hallucination Rates and Reference Accuracy of ChatGPT and Bard for Systematic Reviews: Comparative Analysis
• ChatGPT is bullshit
• Larger and more instructable language models become less reliable
• Measuring short-form factuality in large language models

Cited by

• DPCheatSheet: Using Worked and Erroneous LLM-usage Examples to Scaffold Differential Privacy Implementation
• The trust crisis in artificial intelligence: AI hallucinations and human-AI collaboration

Discussions

• "What’s particularly disconcerting about chatbots is that they can sound so confident when wrong." AI hallucinations can’t be stopped — but these techniques can limit their damage doi.org/10.1038/d41 [bsky, 2 points, 1 comments]