Thursday, Oct 8, 2026 The claims desk. Receipts included. POWERED BY LENZ ↗
IsThis

Language models predict likely next tokens and can produce factual inaccuracies.

The Claim

Generative models predict likely word sequences rather than retrieve facts, which can lead to factual inaccuracies.

Verified by Lenz ·

The Short Version

The claim accurately describes a source of factual errors in text-generating language models. NIST guidance and peer-reviewed research show that predicting likely next tokens can produce factual inaccuracies; generation does not inherently involve retrieving facts from an external source. Retrieval-augmented systems can add that capability, but the claim’s wording correctly says inaccuracies *can* result.

Caveats

  • The evidence primarily addresses text-generating language models, not every type of generative model.
  • Retrieval-augmented systems can consult external sources; retrieval is not excluded from every generative system.
  • The two NIST links refer to the same document and should not be counted as independent sources.

The Receipts

  1. Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile

    tsapps.nist.gov

  2. Evaluating large language models for accuracy incentivizes hallucinations | Nature

    nature.com

  3. Factuality Enhanced Language Models for Open-Ended Text Generation

    papers.nips.cc

  4. Retrieval-augmented generation for natural language processing: a survey | Artificial Intelligence Review | Springer Nature Link

    link.springer.com

  5. TOWARDS UNDERSTANDING FACTUAL KNOWLEDGE OF LARGE LANGUAGE MODELS

    proceedings.iclr.cc

  6. This peer-reviewed article has been accepted for publication but not yet copyedited or typeset, and so may be subject to change during the production process. The article is considered published and may be cited using its DOI. This is an Open Access article, distributed under the terms of the Creative Commons Attribution NonCommercial-NoDerivatives licence (http://creativecommons.org/licenses/by-nc-nd/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is unaltered and is properly cited. The written permission of Cambridge University Press must be obtained for commercial re-use or in order to create a derivative work.

    cambridge.org

  7. arxiv.org

    arxiv.org

  8. [2308.15711] Optimizing Factual Accuracy in Text Generation through Dynamic Knowledge Selection

    ar5iv.labs.arxiv.org

  9. Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile

    nvlpubs.nist.gov

  10. Hallucination Mitigation for Retrieval-Augmented Large ...

    mdpi.com

+ 24 more sources — see the full list on Lenz

Filed Under

Generative Models

More Fact Checks