Still no lie detector for language models: probing empirical and conceptual roadblocks
Open accessThe claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Current lie-detector methods for LLMs are not yet fully reliable, and the probing methods proposed by Burns et al. and Azaria and Mitchell do not adequately generalize.
“Their empirical results suggest that current lie-detector methods for LLMs are not yet fully reliable, and the probing methods proposed by Burns et al. [31] and Azaria and Mitchell [13] do not adequately generalize.”
From A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions (2025) · cites this paper as [163]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
The Framing of Decisions and the Psychology of Choice
Tversky, A., & Kahneman, D. (1981) · Science · doi:10.1126/science.7455683 · Cited by 17,604 · cited 1× · see citing works
“For example, economists have challenged the validity of the famous framing effects of Tversky and Kahneman (1981) by considering the possibility that the subjects in the study updated on higher-order evidence contained in what was and wasn’t said to them, and the rest of the context of the…”
On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜
Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021) · Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency · doi:10.1145/3442188.3445922 · Cited by 6,988 · cited 6× · see citing works
“3.1 we pushed back against the claims in Shanahan (2022) and Bender et al. (2021) that all an LLM is doing is text prediction, and thus cannot be tracking the truth.”
InfluentialThe mythos of model interpretability
Lipton, Z. C. (2018) · Communications of the ACM · doi:10.1145/3233231 · Cited by 5,427 · see citing works
Survey of Hallucination in Natural Language Generation
Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023) · ACM Computing Surveys · doi:10.1145/3571730 · Cited by 5,025 · cited 2× · see citing works
“This is sometimes called lying and sometimes called hallucinating (Ji et al. 2023; Evans et al. 2021).”
Training language models to follow instructions with human feedback
Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C. L., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al. (2022) · arXiv · doi:10.48550/arxiv.2203.02155 · Cited by 4,208 · see citing works
Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data
Bender, E. M., & Koller, A. (2020) · Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2020.acl-main.463 · Cited by 1,082 · cited 2× · see citing works
“There are other worries that LLM belief skeptics bring up, mainly to do with communicative intent and reference (Bender & Koller, 2020; Bender et al., 2021; Shanahan, 2022).”
UNIFIEDQA: Crossing Format Boundaries with a Single QA System
Khashabi, D., Min, S., Khot, T., Sabharwal, A., Tafjord, O., Clark, P., & Hajishirzi, H. (2020) · Findings of the Association for Computational Linguistics: EMNLP 2020 · doi:10.18653/v1/2020.findings-emnlp.171 · Cited by 514 · see citing works
The Internal State of an LLM Knows When It's Lying
Azaria, A., & Mitchell, T. (2023) · arXiv · doi:10.48550/arxiv.2304.13734 · Cited by 179 · see citing works
The Internal State of an LLM Knows When It’s Lying
Azaria, A., & Mitchell, T. (2023) · Findings of the Association for Computational Linguistics: EMNLP 2023 · doi:10.18653/v1/2023.findings-emnlp.68 · Cited by 175 · see citing works
Symbols and grounding in large language models
Pavlick, E. (2023) · Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences · doi:10.1098/rsta.2022.0041 · Cited by 95 · cited 1× · see citing works
“Pavlick (2023) provides a nice high-level summary of both internalist and externalist arguments that LLMs might encode meaning. of a probability function, and desires in the form of a utility function.”
Do Language Models’ Words Refer?
Mandelkern, M., & Linzen, T. (2024) · Computational Linguistics · doi:10.1162/coli_a_00522 · Cited by 77 · see citing works
Discovering Latent Knowledge in Language Models Without Supervision
Burns, C., Ye, H., Klein, D., & Steinhardt, J. (2022) · arXiv · doi:10.48550/arxiv.2212.03827 · Cited by 55 · see citing works
Flagged first, then by citations
Research works printed but not linked: 26
- Alain, G. and Y. Bengio (2016). Understanding intermediate layers using linear classifier probes. arXiv preprint arXiv:1610.01644 . · arXiv
- Cowie, C. (2014). In defence of instrumentalism about epistemic normativity. Synthese 191 (16), 4003-4017.
- Diaconis, P. and B. Skyrms (2018). Ten great ideas about chance . Princeton University Press.
- Evans, O., O. Cotton-Barratt, L. Finnveden, A. Bales, A. Balwit, P. Wills, L. Righetti, and W. Saunders (2021). Truthful ai: Developing and governing ai that does not lie. arXiv preprint arXiv:2110.06674 . · arXiv
- 22 Using latent variables to compute probability distributions is commonplace in science and statistics (Everett (2013)). Though we do not have the space to do latent variable methods full justice, one reason for this is that using distributions over latent variables in order to calculate a distribution over observable variables can have massive computational benefits (see, for example, chapter 16 of Goodfellow et al. (2016)). Thus, it would be fairly surprising if there weren't a useful way to think of LLMs as using some kinds of latent variables in order to make predictions about the next token. Indeed, there is already some preliminary work on what sorts of latent variables LLMs might be working with (Xie et al. (2021); Jiang (2023)).
- Everett, B. (2013). An introduction to latent variable models . Springer Science & Business Media.
- Gilboa, I., S. Minardi, L. Samuelson, and D. Schmeidler (2020). States and contingencies: How to understand savage without anyone being hanged. Revue économique 71 (2), 365-385.
- Gneiting, T. and A. E. Raftery (2007). Strictly proper scoring rules, prediction, and estimation. Journal of the American statistical Association 102 (477), 359-378.
- Godfrey-Smith, P. (1991). Signal, decision, action. The Journal of philosophy 88 (12), 709-722.
- Godfrey-Smith, P. (1998). Complexity and the Function of Mind in Nature . Cambridge University Press.
- Goodfellow, I., Y. Bengio, and A. Courville (2016). Deep learning . MIT press.
- Harding, J. (2023). Operationalising representation in natural language processing. arXiv preprint arXiv:2306.08193 . · arXiv
- Hubinger, E., C. van Merwijk, V. Mikulik, J. Skalse, and S. Garrabrant (2019). Risks from learned optimization in advanced machine learning systems. arXiv preprint arXiv:1906.01820 . · arXiv
- Jeffrey, R. C. (1990). The logic of decision . University of Chicago press.
- Jiang, H. (2023). A latent space theory for emergent abilities in large language models. arXiv preprint arXiv:2304.09960 . · arXiv
- Maas, A., R. E. Daly, P. T. Pham, D. Huang, A. Y. Ng, and C. Potts (2011). Learning word vectors for sentiment analysis. In Proceedings of the 49th annual meeting of the association for computational linguistics: Human language technologies , pp. 142-150.
- Millikan, R. G. (1995). White queen psychology and other essays for Alice . mit Press Cambridge.
- Quine, W. V. (1969). Natural kinds. In Essays in honor of Carl G. Hempel: A tribute on the occasion of his sixty-fifth birthday , pp. 5-23. Springer.
- Quine, W. V. O. (1960). Word and object . MIT Press.
- Shanahan, M. (2022). Talking about large language models. arXiv preprint arXiv:2212.03551 . · arXiv
- Smead, R. (2015). The role of social interaction in the evolution of learning. The British Journal for the Philosophy of Science .
- Sober, E. (1994). The adaptive advantage of learning and a priori prejudice. Ethology and Sociobiology 15 (1), 55-56.
- Stich, S. P. (1990). The fragmentation of reason: Preface to a pragmatic theory of cognitive evaluation. The MIT Press.
- Street, S. (2009). Evolution and the normativity of epistemic reasons. Canadian Journal of Philosophy Supplementary Volume 35 , 213-248.
- Vaswani, A., N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. u. Kaiser, and I. Polosukhin (2017). Attention is all you need. In I. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett (Eds.), Advances in Neural Information Processing Systems , Volume 30. Curran Associates, Inc.
- Xie, S. M., A. Raghunathan, P. Liang, and T. Ma (2021). An explanation of in-context learning as implicit bayesian inference. arXiv preprint arXiv:2111.02080 . · arXiv
Cited by
Sorted by citations
Huang, L., Yu, W., Ma, W., Zhong, W., Feng, Z., Wang, H., Chen, Q., Peng, W., Feng, X., Qin, B., et al. (2025) · ACM Transactions on Information Systems · doi:10.1145/3703155 · Cited by 2,142 · see citing works
“Nonetheless, Levinstein and Herrmann (2023) have employed empirical and conceptual tools to probe whether or not LLMs have beliefs.”
Hicks, M. T., Humphries, J., & Slater, J. (2024) · Ethics and Information Technology · doi:10.1007/s10676-024-09775-5 · Cited by 456 · see citing works
AI Deception: A Survey of Examples, Risks, and Potential Solutions
Park, P. S., Goldstein, S., O'Gara, A., Chen, M., & Hendrycks, D. (2023) · arXiv · doi:10.48550/arxiv.2308.14752 · Cited by 243 · see citing works
“For example, Levinstein et al. 91 have raised reliability issues for the strategies of Azaria et al. and Burns et al., although these reliability issues have recently been mitigated by the strategy of Zou et al. 92 Also, Pacchiardi et al. 93 introduced a lie-de-tector method that asks a collection…”
AI deception: A survey of examples, risks, and potential solutions
Park, P. S., Goldstein, S., O’Gara, A., Chen, M., & Hendrycks, D. (2024) · Patterns · doi:10.1016/j.patter.2024.100988 · Cited by 243 · see citing works
“For example, Levinstein et al. 91 have raised reliability issues for the strategies of Azaria et al. and Burns et al., although these reliability issues have recently been mitigated by the strategy of Zou et al. 92 Also, Pacchiardi et al. 93 introduced a lie-de-tector method that asks a collection…”
Huang, L., Yu, W., Ma, W., Zhong, W., Feng, Z., Wang, H., Chen, Q., Peng, W., Feng, X., Qin, B., et al. (2023) · arXiv · doi:10.48550/arxiv.2311.05232 · Cited by 2,516 · see citing works
“Nonetheless, Levinstein and Herrmann (2023) have employed empirical and conceptual tools to probe whether or not LLMs have beliefs.”
Why we need to be careful with LLMs in medicine
Bélisle-Pipon, J.-C. (2024) · Frontiers in Medicine · doi:10.3389/fmed.2024.1495582 · Cited by 91 · see citing works
A Survey of Confidence Estimation and Calibration in Large Language Models . -
Geng, J., Cai, F., Wang, Y., Koeppl, H., Nakov, P., & Gurevych, I. (2024) · Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) · doi:10.18653/v1/2024.naacl-long.366 · Cited by 88 · see citing works
From Illusion to Insight: A Taxonomic Survey of Hallucination Mitigation Techniques in LLMs
Kazlaris, I., Antoniou, E., Diamantaras, K., & Bratsas, C. (2025) · doi:10.20944/preprints202508.1942.v1 · Cited by 13 · see citing works
Multi-Agent Risks from Advanced AI
Hammond, L., Chan, A., Clifton, J., Hoelscher-Obermaier, J., Khan, A., McLean, E., Smith, C., Barfuss, W., Foerster, J., Gavenčiak, T., et al. (2025) · arXiv · doi:10.48550/arxiv.2502.14143 · Cited by 10 · see citing works
- Authors
- Levinstein, B. A., & Herrmann, D. A.
- Affiliations
- University of Illinois Urbana-Champaign (US)
- University of Groningen (NL)
- Corresponding author
- Levinstein, Benjamin A.
- Venue
- Philosophical Studies
- Published
- 2024-02-17
- Volume
- 182
- Issue
- 7
- Pages
- 1539-1565
- ISSN
- 0031-8116 (ISSN-L)
- 1573-0883
- OpenAlex
- W4391914041
- Cited by
- 50
- References
- 12
- Citations by year
- 2026: 21 · 2025: 16 · 2024: 11 · 2023: 2
- FWCI
- 19.73
- Type
- journal article
- Language
- en
- Open access
- green
- research.rug.nl
- Licence
- Locations
- Philosophical Studies (publishedVersion)
- University of Groningen research database (University of Groningen / Centre for Information Technology) (publishedVersion, other-oa)
- arXiv (Cornell University) (submittedVersion, cc-by-nc-sa)
- arXiv (Cornell University)
- Funders
- Andrew W. Mellon Foundation (1905-06835)
- Open Philanthropy Project
- Long-Term Future Fund
- Topics
- Social Sciences › Psychology › Experimental and Cognitive Psychology › Philosophy and Theoretical Science
- Social Sciences › Arts and Humanities › Philosophy › Epistemology, Ethics, and Metaphysics
- Physical Sciences › Computer Science › Artificial Intelligence › Logic, Reasoning, and Knowledge
- Keywords
- large language models, belief, lying, deception detection, internal state, belief measurement, generalization, philosophy of artificial intelligence
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
