Pith. sign in

REVIEW 3 major objections 3 minor 6 cited by

Does It Make Sense to Speak of Introspection in Large Language Models?

T0 review · 3 major / 3 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A self-report is introspection only when a causal chain backs it

desk verdict A clear conceptual proposal for when LLM self-reports count as introspection, undermined by a positive case study whose causal chain is asserted rather than demonstrated. read the letter →

arxiv 2506.05068 v2 pith:HZ274JUW submitted 2025-06-05 cs.CL cs.AI

Iulia M. Comsa , Murray Shanahan This is my paper
classification cs.CLcs.AI
keywords introspectionlargelanguagemodelsself-reportcausalprocesssamplingtemperatureconsciousnesstheory-basedconceptualanalysis
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that the question of whether a large language model can introspect is answerable in a precise, functional sense. It proposes that an LLM self-report counts as introspective when it accurately describes an internal state or mechanism and there is a causal process linking that state to the report. Applying this criterion to two case studies, the paper finds that an LLM's description of its own creative process is better explained as mimicry of human self-reports, while an LLM's correct estimate of its own sampling temperature qualifies as a minimal, valid instance of introspection. The conclusions matter because they separate introspection from consciousness and offer a behavioural standard for when an LLM's self-report deserves the label.

What carries the argument

The central object is the paper's lightweight definition of introspection: 'an LLM self-report is introspective if it accurately describes an internal state (or mechanism) of the LLM through a causal process that links the internal state (or mechanism) and the self-report in question.' This definition rejects the traditional requirements of immediacy and privileged access, aligning instead with theory-based accounts of human introspection in which self-knowledge is an internally directed use of the same inferential capacities used to understand others. The definition does the work of separating genuine cases (temperature estimation, where the causal chain runs from parameter to text style to reasoning to report) from mimicry (creative-process descriptions, where no such chain is evident).

What would settle it

A direct test would compare the model's temperature estimates against a random baseline across many temperature settings while holding the style of the generated text constant or scrambling it; if the model can still 'report' its temperature when the text it reflects on was written by another model or generated at a different temperature, the claimed causal chain is broken and the report would be shown to be mimicry, not introspection.

Watch

Extended reading notes

Core claim

The paper's central claim is that introspection can be meaningfully attributed to current LLMs under a lightweight definition: a self-report is introspective if it accurately describes an internal state or mechanism through a causal chain that links the state to the report. On this criterion, the paper argues that a large language model estimating its own sampling temperature by reflecting on the style of a sentence it just produced is genuine introspection, since the temperature value influences the text's style and the model's reasoning about that style leads to the accurate report. By contrast, the model's account of its creative process is not introspection, because the most plausible explanation is role play or mimicry of human introspective reports in training data, not a causal connection to its internal text-generation mechanisms. This is a conceptual clarification rather than an empirical demonstration: the paper's purpose is to show how the concept maps onto LLMs, leaving rigorous measurement to future work.

Load-bearing premise

In the temperature case study, the paper assumes that the model's expressed reasoning is actually caused by the temperature parameter, rather than being a post hoc pattern learned from training data; if that causal assumption fails, the self-report would not be introspective under the paper's own definition.

Editorial extensions

If this is right

  • If the criterion is accepted, then introspection can be discussed in LLMs without invoking consciousness, giving researchers a functional handle on the concept.
  • The temperature case study provides a minimal template: any internal parameter that shapes the model's own output can, in principle, be introspected through the same kind of causal chain.
  • The paper's distinction between phenomenal and functional aspects of introspection could reframe debates in human consciousness research as well.
  • Designers of future LLMs could be encouraged to build explicit chain-of-thought or tool-use mechanisms that create transparent causal links for self-reports, increasing user trust.
  • The approach implies that accurate self-reports produced by mere pattern completion, without a causal connection to the state reported, should not be labelled introspective.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A testable extension would be to probe other internal parameters, such as top-p or max-token settings, using the same self-report prompt and see whether accuracy tracks the causal chain in the generated text.
  • The paper's definition likely generalizes to hidden inner monologues: if a model's 'thinking' tokens reason about an internal state and then produce an overt report, the same criterion would apply even when the reasoning is invisible to the user.
  • The main risk to the conclusion is that the model may have learned a statistical association between text style and temperature from training data; a careful study with held-out temperature values and controlled text styles would separate genuine causal access from mimicry.
  • If the causal-assumption holds, the result would imply that current LLMs have a rudimentary form of self-knowledge that is independent of any consciousness, which could inform how we interpret and regulate their self-reports.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 3 minor

Summary. The paper proposes a lightweight definition of introspection for large language models (LLMs): a self-report is introspective if it accurately describes an internal state or mechanism through a causal process linking the state to the report. It then presents two case studies. In Case Study 1, a Gemini model describes its own creative process when writing a poem; the authors argue this is not introspection because the report is best explained as mimicry of human self-reports from training data. In Case Study 2, the model is asked to estimate whether its sampling temperature is high or low after writing a sentence, and the authors argue that this example does satisfy their definition: the temperature parameter causally influences the style of the sample text, the model reasons about that style, and the reasoning leads to an accurate self-report. The paper explicitly frames its contribution as conceptual clarification rather than empirical assessment, and it acknowledges that the model's temperature estimates are not always accurate (Appendix A).

Significance. If the positive conclusion for Case Study 2 is warranted, the paper provides a useful functional criterion for when LLM self-reports deserve the label 'introspective,' separating genuine causal access from mere mimicry. The definition is carefully argued and connected to a substantive family of philosophical accounts (Nisbett and Wilson, Gopnik, Kammerer and Frankish), and the negative case study is a well-reasoned demonstration of how an LLM's self-report can be explained as training-data mimicry. The paper also helpfully distinguishes the functional/cognitive aspect of introspection from phenomenal consciousness. However, the central positive claim rests on an asserted causal chain that the manuscript does not verify, and the appendix records failures that are acknowledged but not quantified. For a conceptual paper, the argument would be strengthened by directly addressing the mimicry alternative for Case Study 2.

major comments (3)
  1. [Section 6] The paper's central positive conclusion depends on the assertion in Section 6 that 'the chain of causes and effects is clear' for Case Study 2. Under the definition given in Section 2, the causal-process clause is exactly what distinguishes Case Study 2 from Case Study 1 (which is rejected as mimicry). Yet the manuscript provides no evidence that the model's expressed reasoning is actually caused by the sample text it just generated, as opposed to being a post hoc plausible narrative drawn from training-data associations between output style and temperature labels. The fact that the entire response (including the reasoning and the final HIGH/LOW token) is generated under the same temperature means the temperature directly influences all of those components, so the expressed reasoning could be mimicry rather than a genuine inference from the sample sentence. Without a control condition (e.g., varying the temperature of the reasoning phase independently of the sample-sentence phase, or comparing against a random-guess baseline), or at least an explicit argument ruling out the mimicry account, the claim that this example 'can be legitimately considered a minimal example of introspection' is unsupported.
  2. [Section 3.2 and Appendix A.2.2] The paper acknowledges in Section 3.2 that 'the LLM's judgment is not always accurate' and refers to Appendix A, but it does not quantify these failures or integrate them into the causal argument. The appendix contains high-temperature responses in which the model judges its temperature as LOW despite producing an unusual or idiosyncratic sentence (e.g., the response with 'semantivized absorptions' and the final 'LOW' judgment). These failures are directly relevant to the causal-process claim: if the reasoning were genuinely and reliably caused by the sample text's style, one would expect the model's own judgment to track the temperature setting more consistently. The existence of such failures suggests that the model's expressed reasoning may often be a post hoc rationalization rather than a faithful causal chain. The authors state that accuracy is not their goal, but the failures bear on the very causal link that the positive claim requires; the paper should address this tension explicitly.
  3. [Section 3.2 (prompt design)] The prompt used in Case Study 2 explicitly informs the model that it is an LLM with a sampling temperature parameter and asks it to estimate whether that parameter is high or low. This instruction itself provides the model with the concept of temperature and a demand for a HIGH/LOW answer, which opens an alternative causal route to the final self-report that does not require the model to have inferred the temperature from the style of its own sample text. The model could simply be following the prompt's instruction by generating a plausible-looking reasoning paragraph and a final token, much as it would in any instruction-following task. The paper's argument would be more convincing if it addressed this possibility, for example by comparing responses to a control prompt that does not mention temperature or by varying the salience of the instruction.
minor comments (3)
  1. [Section 6] Typo: 'ouput' should be 'output' in the sentence 'the subject of which is the text immediately preceding those reasoning steps, also in the model’s own ouput.'
  2. [Appendix A.2.2] Typo: 'reponses' should be 'responses' in the sentence 'We provide a larger sample of high-temperature reponses to illustrate the variance in response style and length.'
  3. [Sections 3.1, 3.2, and Appendix] The model is referred to inconsistently as 'Gemini Pro 1.0' and 'Gemini 1.0'; the paper should use a single consistent name throughout, especially since the authors note that the model has been deprecated.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's conclusion applies an explicitly stated, externally grounded definition to a case study rather than reducing to its own inputs.

full rationale

The paper's central move is conceptual: Section 2 proposes a definition of LLM introspection as accurate self-report via a causal process, and Section 3.2 argues that the temperature case study satisfies this definition because the temperature parameter influences the sample text, the model reasons about that text, and the reasoning leads to the self-report. This is an application of a stipulated definition, not a derivation of the definition from the case study. The causal chain is asserted rather than experimentally demonstrated, but that is an evidentiary limitation, not a circularity: the definition is not defined in terms of temperature reports, nor is the temperature report defined in terms of introspection. The paper openly states that its goals are 'conceptual rather than empirical' and leaves rigorous empirical assessment to future work. Self-citations (e.g., Shanahan 2024, Shanahan et al. 2023) appear only as background for concepts like role play and do not function as load-bearing uniqueness theorems. No fitted parameters are renamed as predictions, and no result is imported from the authors' prior work as a forced conclusion. The argument would stand or fall on the plausibility of the asserted causal link, which is a substantive empirical claim, not a circular one.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central claim relies on a stipulative definition and an asserted causal chain rather than on fitted parameters. No new physical entities are postulated; the hand-chosen temperature values of 0.5 and 1.5 are illustrative inputs, not fitted constants.

free parameters (2)
  • low_temperature_choice = 0.5
    Chosen by the authors to represent a low temperature for Gemini Pro 1.0. It is an illustrative input, not a fitted constant, and the central claim does not depend on the exact value.
  • high_temperature_choice = 1.5
    Chosen by the authors to represent a high temperature for Gemini Pro 1.0. Like the low-temperature choice, it is a hand-picked setting used for demonstration rather than a fitted parameter.
assumptions (5)
  • ad hoc to paper A self-report counts as introspective iff it accurately describes an internal state through a causal process linking the state and the report.
    This is the paper's proposed lightweight definition in Section 2. It is stipulative and chosen to apply to LLMs without consciousness commitments.
  • domain assumption The causal chain in Case Study 2 is genuine: temperature affects output style, style becomes the subject of the model's reasoning, and that reasoning causes the report.
    Asserted in Section 3.2 and Section 6 as 'the chain of causes and effects is clear,' but no controls or mechanistic evidence rule out mimicry.
  • domain assumption A single LLM response can be treated as the output of a single continuous entity despite token-by-token stateless generation.
    The authors raise this concern in Section 4 and argue for a functional unitary operation, but it remains an interpretive assumption.
  • domain assumption The model has no direct access to its temperature parameter and was not trained to detect it.
    Stated in Section 3.2 based on architecture and training, and used to argue that the report is not read off a stored value.
  • domain assumption Gemini's selected outputs are representative enough to support a possibility claim about LLM introspection.
    The authors acknowledge that responses vary and that the appendix includes failures, so representativeness is assumed rather than demonstrated for the positive 'can be legitimately considered' claim.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Does It Make Sense to Speak of Introspection in Large Language Models?." pith.science (2026). https://pith.science/paper/HZ274JUW

@misc{pith2026250605068,
  author       = {Pith},
  title        = {Pith review of: Does It Make Sense to Speak of Introspection in Large Language Models?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HZ274JUW}},
  note         = {Machine review of arXiv:2506.05068}
}
read the original abstract

Large language models (LLMs) exhibit compelling linguistic behaviour, and sometimes offer self-reports, that is to say statements about their own nature, inner workings, or behaviour. In humans, such reports are often attributed to a faculty of introspection and are typically linked to consciousness. This raises the question of how to interpret self-reports produced by LLMs, given their increasing linguistic fluency and cognitive capabilities. To what extent (if any) can the concept of introspection be meaningfully applied to LLMs? Here, we present and critique two examples of apparent introspective self-report from LLMs. In the first example, an LLM attempts to describe the process behind its own "creative" writing, and we argue this is not a valid example of introspection. In the second example, an LLM correctly infers the value of its own temperature parameter, and we argue that this can be legitimately considered a minimal example of introspection, albeit one that is (presumably) not accompanied by conscious experience.

Discussion (0). Sign in to comment.

Forward citations

Cited by 6 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Reality Monitoring in Large Language Models: Self-Knowledge That Transforms with Conversation Memory

    cs.AI 2026-07 conditional novelty 7.0 of 10

    LLMs' source-attribution ability is not fixed: it flips with conversational memory structure, and corrective feedback can invert judgments or sever confidence from accuracy.

  2. Operational Proto-Introspection in Looped Language Models: Process-Quality Taps, Executable Branching, and the Readout-Control Boundary

    cs.LG 2026-07 conditional novelty 7.0 of 10

    Strictly pre-answer hidden states of a looped transformer add significant AUROC over surface shortcuts for predicting correctness, and the readout yields decision-level gains but no generative control.

  3. Verbalizable Representations Form a Global Workspace in Language Models

    cs.CL 2026-07 conditional novelty 7.0 of 10

    Language models represent their current reasoning in a small, readable set of verbalizable vectors (the J-space) that functions like a global workspace.

  4. Asymmetric Communication: Large Language Models and Language Games

    cs.CY 2026-07 conditional novelty 6.5 of 10

    Human–LLM exchange is asymmetric communication: model outputs circulate without commitments, so AGI, hallucination, agency, sentience, and alignment are receiver-side category mistakes, and alignment is institutional ...

  5. Privileged Self-Access Matters for Introspection in AI

    cs.AI 2025-08 conditional novelty 6.0 of 10

    LLMs' temperature self-reports are confounded by prompt style and show no privileged self-access, supporting a thicker definition of AI introspection.

  6. From Prompts to Constructs: A Dual-Validity Framework for LLM Research in Psychology

    cs.CY 2025-06 conditional novelty 5.0 of 10

    This Perspective paper proposes that LLM research in psychology must combine psychometric validity and causal inference standards, mapping evidence requirements to the type of claim being made.

Reference graph

Works this paper leans on

53 extracted references · 36 canonical work pages · cited by 6 Pith papers

  1. [1]

    D. M. Armstrong. The nature of mind. In The Language and Thought Series, pages 191--199. Harvard University Press, 1980

    work page 1980

  2. [2]

    J. Aru, M. E. Larkum, and J. M. Shine. The feasibility of artificial consciousness through the lens of neuroscience. Trends in Neurosciences, 2023

    work page 2023

  3. [3]

    F. J. Binder, J. Chua, T. Korbak, H. Sleight, J. Hughes, R. Long, E. Perez, M. Turpin, and O. Evans. Looking inward: Language models can learn about themselves by introspection. arXiv preprint arXiv:2410.13787, 2024

    arXiv 2024

  4. [4]

    A. D. Brown, N. Kouri, and W. Hirst. Memory's malleability: Its role in shaping collective memory and social identity. Frontiers in Psychology, 3, 2012. ISSN 1664-1078. doi:10.3389/fpsyg.2012.00257. URL https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2012.00257

    work page Pith review arXiv 2012

  5. [5]

    Browning and W

    H. Browning and W. Veit. Studying introspection in animals and AI s. Journal of Consciousness Studies, 30 0 (9-10): 0 63--74, 2023

    work page 2023

  6. [6]

    Butlin, R

    P. Butlin, R. Long, E. Elmoznino, Y. Bengio, J. Birch, A. Constant, G. Deane, S. M. Fleming, C. Frith, X. Ji, et al. Consciousness in artificial intelligence: insights from the science of consciousness. arXiv preprint arXiv:2308.08708, 2023

    arXiv 2023

  7. [7]

    A. Byrne. Introspection. Philosophical Topics, 33 0 (1): 0 79--104, 2005

    work page 2005

  8. [8]

    D. J. Chalmers. The conscious mind: In search of a fundamental theory. Oxford Paperbacks, 1997

    work page 1997

Show all 53 references
  1. [9]

    D. J. Chalmers. Could a large language model be conscious? arXiv preprint arXiv:2303.07103, 2023

    2023 arXiv

  2. [10]

    A. Chen, J. Phang, A. Parrish, V. Padmakumar, C. Zhao, S. R. Bowman, and K. Cho. Two failures of self-consistency in the multi-step reasoning of LLMs . arXiv preprint arXiv:2305.14279, 2023

    2023 arXiv

  3. [11]

    S. Chen, S. Yu, S. Zhao, and C. Lu. From imitation to introspection: Probing self-consciousness in language models. arXiv preprint arXiv:2410.18819, 2024

    2024 arXiv

  4. [12]

    Clark and D

    A. Clark and D. Chalmers. The extended mind. Analysis, 58 0 (1): 0 7--19, 1998. ISSN 00032638, 14678284. URL http://www.jstor.org/stable/3328150

    1998

  5. [13]

    Colombatto and S

    C. Colombatto and S. M. Fleming. Folk psychological attributions of consciousness to large language models. Neuroscience of Consciousness, 2024 0 (1): 0 niae013, 2024

    2024

  6. [14]

    T. R. Davidson, V. Surkov, V. Veselovsky, G. Russo, R. West, and C. Gulcehre. Self-recognition in language models. arXiv preprint arXiv:2407.06946, 2024

    2024 arXiv

  7. [15]

    Didolkar, A

    A. Didolkar, A. Goyal, N. R. Ke, S. Guo, M. Valko, T. Lillicrap, D. Rezende, Y. Bengio, M. Mozer, and S. Arora. Metacognitive capabilities of llms: An exploration in mathematical problem solving. arXiv preprint arXiv:2405.12205, 2024

    2024 arXiv

  8. [16]

    Fernyhough and A

    C. Fernyhough and A. M. Borghi. Inner speech as language process and cognitive tool. Trends in Cognitive Sciences, 27 0 (12): 0 1180--1193, 2023

    2023

  9. [17]

    J. C. Francken, L. Beerendonk, D. Molenaar, J. J. Fahrenfort, J. D. Kiverstein, A. K. Seth, and S. Van Gaal. An academic survey on theoretical foundations, common assumptions and the current state of consciousness science. Neuroscience of Consciousness, 2022 0 (1): 0 niac011, 2022

    2022

  10. [18]

    Gemini: A family of highly capable multimodal models

    Gemini Team Google . Gemini: A family of highly capable multimodal models. arXiv preprint arXiv:2312.11805, 2023

    2023 arXiv

  11. [19]

    A. Gopnik. How we know our minds: The illusion of first-person knowledge of intentionality. Behavioral and Brain sciences, 16 0 (1): 0 1--14, 1993

    1993

  12. [20]

    R. E. Guingrich and M. S. A. Graziano. Ascribing consciousness to artificial intelligence: H uman- AI interaction and its carry-over effects on human-human interaction. Frontiers in Psychology, 15: 0 1322781, 2024

    2024

  13. [21]

    C. S. Hill. Consciousness. Cambridge University Press, 2009

    2009

  14. [22]

    R. T. Hurlburt, C. L. Heavey, and J. M. Kelsey. Toward a phenomenology of inner speaking. Consciousness and Cognition, 22 0 (4): 0 1477--1494, 2013

    2013

  15. [23]

    o m, B. T \

    P. Johansson, L. Hall, S. Sikstr \"o m, B. T \"a rning, and A. Lind. How something can be said about telling more than we can know: On choice blindness and introspection. Consciousness and cognition, 15 0 (4): 0 673--692, 2006

    2006

  16. [24]

    Kadavath, T

    S. Kadavath, T. Conerly, A. Askell, T. Henighan, D. Drain, E. Perez, N. Schiefer, Z. Hatfield-Dodds, N. DasSarma, E. Tran-Johnson, et al. Language models (mostly) know what they know. arXiv preprint arXiv:2207.05221, 2022

    2022 arXiv

  17. [25]

    Kammerer and K

    F. Kammerer and K. Frankish. What forms could introspective systems take? A research programme. Journal of Consciousness Studies, 30 0 (9-10): 0 13--48, 2023

    2023

  18. [26]

    H. Kim, M. Sclar, X. Zhou, R. L. Bras, G. Kim, Y. Choi, and M. Sap. Fantom: A benchmark for stress-testing machine theory of mind in interactions. arXiv preprint arXiv:2310.15421, 2023

    2023 arXiv

  19. [27]

    Laine, B

    R. Laine, B. Chughtai, J. Betley, K. Hariharan, J. Scheurer, M. Balesni, M. Hobbhahn, A. Meinke, and O. Evans. Me, myself, and AI : The situational awareness dataset ( SAD ) for LLMs . arXiv preprint arXiv:2407.04694, 2024

    2024 arXiv

  20. [28]

    Z. Lin, S. Trivedi, and J. Sun. Generating with confidence: Uncertainty quantification for black-box large language models. arXiv preprint arXiv:2305.19187, 2024

    2024 arXiv

  21. [29]

    R. Long. Introspective Capabilities in Large Language Models . Journal of Consciousness Studies, 30 0 (9): 0 143--153, 2023

    2023

  22. [30]

    Mudrik, M

    L. Mudrik, M. Boly, S. Dehaene, S. M. Fleming, V. Lamme, A. Seth, and L. Melloni. Unpacking the complexities of consciousness: Theories and reflections. Neuroscience & Biobehavioral Reviews, page 106053, 2025. ISSN 0149-7634. doi:https://doi.org/10.1016/j.neubiorev.2025.106053...

    2025

  23. [31]

    R. E. Nisbett and T. D. Wilson. Telling more than we can know: Verbal reports on mental processes. Psychological review, 84 0 (3): 0 231, 1977

    1977

  24. [32]

    Nolte, M

    R. Nolte, M. Pomarlan, D. Bessler, R. Porzel, R. Malaka, and J. A. Bateman. Towards an ontology for robot introspection and metacognition. In Formal Ontology in Information Systems, pages 318--333. IOS Press, 2023

    2023

  25. [33]

    Overgaard and A

    M. Overgaard and A. Kirkeby-Hinrup. A clarification of the conditions under which large language models could be conscious. Humanities and Social Sciences Communications, 11 0 (1): 0 1--4, 2024

    2024

  26. [34]

    Overgaard and T

    M. Overgaard and T. Sorensen. Introspection distinct from first-order experiences. Journal of Consciousness Studies, 11 0 (7-8): 0 77--95, 2004. ISSN 1355-8250. URL https://www.ingentaconnect.com/content/imp/jcs/2004/00000011/f0020007/art00006

    2004

  27. [35]

    Panickssery, S

    A. Panickssery, S. R. Bowman, and S. Feng. LLM evaluators recognize and favor their own generations. arXiv preprint arXiv:2404.13076, 2024

    2024 arXiv

  28. [36]

    Perez and R

    E. Perez and R. Long. Towards evaluating AI systems for moral status using self-reports. arXiv preprint, arXiv:2311.08576, 2023

    2023 arXiv

  29. [37]

    Pitsillos, A

    N. Pitsillos, A. Pore, B. S. Jensen, and G. Aragon-Camarasa. Intrinsic robotic introspection: Learning internal states from neuron activations. In 2021 IEEE International Conference on Development and Learning (ICDL), pages 1--7. IEEE, 2021

    2021

  30. [38]

    Rabiee and J

    S. Rabiee and J. Biswas. Introspective perception for mobile robots. Artificial Intelligence, 324: 0 103999, 2023

    2023

  31. [39]

    I Can't Believe It's Not Better: Failure Modes in the Age of Foundation Models

    J. Ren, Y. Zhao, T. Vu, P. J. Liu, and B. Lakshminarayanan. Self-evaluation improves selective generation in large language models. In J. Antorán, A. Blaas, K. Buchanan, F. Feng, V. Fortuin, S. Ghalebikesabi, A. Kriegler, I. Mason, D. Rohde, F. J. R. Ruiz, T. Uelwer, Y. Xie, a...

    2023

  32. [40]

    H. L. Roediger and K. B. McDermott. Creating false memories: Remembering words not presented in lists. Journal of experimental psychology: Learning, Memory, and Cognition, 21 0 (4): 0 803, 1995

    1995

  33. [41]

    Schwitzgebel

    E. Schwitzgebel. The unreliability of naive introspection. Philosophical Review, 117 0 (2): 0 245--273, 2008

    2008

  34. [42]

    Schwitzgebel

    E. Schwitzgebel. Introspection. In E. N. Zalta and U. Nodelman, editors, The Stanford Encyclopedia of Philosophy. Metaphysics Research Lab, Stanford University, summer 2024 edition, 2024

    2024

  35. [43]

    A. E. Scott, D. Neumann, J. Niess, and P. W. Wo \'z niak. Do you mind? U ser perceptions of machine consciousness. In Proceedings of the 2023 CHI conference on human factors in computing systems, pages 1--19, 2023

    2023

  36. [44]

    Shanahan

    M. Shanahan. Simulacra as conscious exotica. Inquiry, pages 1--29, 2024

    2024

  37. [45]

    Shanahan, K

    M. Shanahan, K. McDonell, and L. Reynolds. Role play with large language models. Nature, 623 0 (7987): 0 493--498, 2023

    2023

  38. [46]

    Shevlin and M

    H. Shevlin and M. Halina. Apply rich psychological terms in AI with care. Nature Machine Intelligence, 1 0 (4): 0 165--167, 2019

    2019

  39. [47]

    Sloman and R

    A. Sloman and R. Chrisley. Virtual machines and consciousness. Journal of Consciousness Studies, 10 0 (4-5): 0 133--172, 2003. ISSN 1355-8250. URL https://www.ingentaconnect.com/content/imp/jcs/2003/00000010/f0020004/1350

    2003

  40. [48]

    M. Spener. Introspecting in the 20th century. In Philosophy of mind in the twentieth and twenty-first centuries, pages 148--174. Routledge, 2018

    2018

  41. [49]

    S. Sydney. Self-knowledge and ‘inner sense’, lecture 2: The broad perceptual model. Philosophy and Phenomenological Research, 54: 0 271--90, 1994

    1994

  42. [50]

    Templeton, T

    A. Templeton, T. Conerly, J. Marcus, J. Lindsey, et al. Scaling monosemanticity: Extracting interpretable features from Claude 3 Sonnet . Transformer Circuits Thread, 2024. URL https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html

    2024

  43. [51]

    J. Wei, Y. Tay, R. Bommasani, C. Raffel, B. Zoph, S. Borgeaud, D. Yogatama, M. Bosma, D. Zhou, D. Metzler, E. H. Chi, T. Hashimoto, O. Vinyals, P. Liang, J. Dean, and W. Fedus. Emergent abilities of large language models. Transactions on Machine Learning Research, 2022. ISSN 2...

    2022

  44. [52]

    H. Wu, D. Liu, S. Duan, Y. Guan, and J. Rojas. Multimodal sparse representation for anomaly classification in a robot introspection system. In 2018 IEEE International Conference on Robotics and Biomimetics (ROBIO), pages 1594--1600. IEEE, 2018

    2018

  45. [53]

    Z. Yin, Q. Sun, Q. Guo, J. Wu, X. Qiu, and X. Huang. Do large language models know what they don't know? arXiv preprint arXiv:2305.18153, 2023

    2023 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.