Pith. sign in

REVIEW 3 major objections 4 minor 38 references

Extending Epistemic Uncertainty Beyond Parameters Would Assist in Designing Reliable LLMs

T0 review · 3 major / 4 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 Bayesian Modeling of Experiments, applied to auxiliary context, can turn LLM uncertainty from a reason to abstain into an actionable signal for clarification, retrieval, and prompt refinement.

desk verdict A clean and honest position paper that mislabels output sensitivity as reducible epistemic uncertainty; worth reviewing but needs a major scope fix. read the letter →

arxiv 2506.07448 v1 pith:2DRBPSBJ submitted 2025-06-09 cs.LG cs.AI

T. Duy Nguyen-Hien , Desi R. Ivanova , Yee Whye Teh , Wee Sun Lee This is my paper
classification cs.LGcs.AI
keywords BayesianModelingofExperimentsepistemicuncertaintyLLMreliabilityexpectedinformationgainclarificationpromptengineeringin-contextlearningdecomposition
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 position paper argues that the usual practice of treating LLM uncertainty as a total score and rejecting uncertain outputs misses what is actually reducible. It proposes Bayesian Modeling of Experiments (BME) as a framework in which epistemic uncertainty is, by definition, uncertainty that can be reduced by running an experiment—here, providing a clarification, a different prompt template, or a set of in-context examples. Under this view, the paper claims, several existing heuristics for improving LLM outputs are the same quantity: expected information gain, measured as a drop in predictive entropy. A sympathetic reader would care because, if true, building reliable LLM systems becomes a matter of choosing which experiment to run, not just when to stay silent.

What carries the argument

The central object is Bayesian Modeling of Experiments (BME), the classical scheme in which an experiment is a random variable $\Lambda$, prior knowledge is a distribution over a quantity of interest $\Xi$, and the value of an experiment is its expected information gain, $H[P(\Xi)] - \mathbb{E}_{P(\Lambda)} H[P(\Xi|\lambda)]$, which equals the mutual information $I[\Xi;\Lambda]$ and the KL divergence between joint and product distributions. In the LLM setting the paper applies this with the predictive distribution $P(Y|X)$ as the quantity of interest and auxiliary context $C$ as the experiment stimulus, giving $I[Y|X;C] = H[P(Y|X)] - \mathbb{E}_{P(C|X)} H[P(Y|X,C)]$. The framework does the work of separating reducible from effectively irreducible uncertainty: what counts as reducible depends on which experiment apparatus is available, and the same entropy-difference identity makes clarification, template choice, and in-context demonstrations instances of one mechanism.

What would settle it

Fix a set of ambiguous user requests, compute the BME score $I[Y;C|X]$ using the LLM's own sampled clarifications, then collect human clarifications and measure the actual reduction in response entropy or error rate for each request. If the ordering of requests by predicted information gain does not match the ordering by actual improvement—say, a near-zero rank correlation—then the framework's central promise that this quantity identifies reducible uncertainty fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that epistemic uncertainty in LLMs should be extended beyond uncertainty about model parameters to any uncertainty reducible by an experiment, and that the Bayesian Modeling of Experiments framework supplies the right quantity for this: the mutual information $I[Y|X;C] = H[P(Y|X)] - \mathbb{E}_{P(C|X)} H[P(Y|X,C)]$, where $C$ is auxiliary context such as a clarifying question, prompt template, or in-context example. It asserts this quantity unifies the otherwise ad hoc treatments of uncertainty in clarification ensembling, prompt selection, and few-shot demonstration selection, and that it resolves apparent contradictions in prior work where reducible uncertainty was mislabeled as aleatoric. The paper further claims BME supports joint management of multiple uncertainty sources and yields a lower bound on effectively irreducible uncertainty, which can justify conservative abstention when the bound is high.

Load-bearing premise

The framework's useful outputs depend on the LLM being a reliable simulator of the alternative contexts and responses it has not actually observed, so that the information-gain estimates reflect real rather than imagined improvements; the paper itself notes such simulators are not always available or reliable.

Editorial extensions

If this is right

  • If BME is right, uncertainty estimates should be reported per source—ambiguity, template, demonstrations—rather than as one total score.
  • Clarification systems can stop approximating ambiguity heuristically and instead pick the question with maximum expected information gain.
  • Prompt engineering and in-context example selection become optimization of the same mutual-information objective, allowing shared methods across tasks.
  • The distinction between epistemic and aleatoric uncertainty becomes relative to the available experiments, yielding an 'effective aleatoric uncertainty' bound that can govern when to abstain.
  • LLMs that can simulate clarifications can bypass arbitrary priors for sources inside their knowledge, making the framework applicable without hand-specified priors.

Reading between the lines

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

  • The paper leaves implicit a cost-aware version: when clarification, web retrieval, and re-prompting have different costs, the rational policy is to run the experiment with the highest expected information gain per unit cost, not merely the highest gain.
  • A natural test is to rank ambiguous prompts by predicted $I[Y;C|X]$ and check whether human-issued clarifications improve response accuracy in that order; the paper motivates but does not run such a validation.
  • The effective-aleatoric lower bound could plug directly into risk-controlled abstention rules, giving a principled middle ground between passive rejection and active clarification.
  • For LLM agents, the same identity suggests a general operating principle: treat tool calls and environment queries as experiments and choose actions that maximize expected reduction in the agent's predictive entropy about the task.
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 / 4 minor

Summary. The paper is a position paper proposing that Bayesian Modeling of Experiments (BME) be adopted as the conceptual foundation for managing uncertainty in LLM deployments. It reviews the BME framework, defines an information-theoretic quantity I[Y|X;C] intended to capture uncertainty reducible by auxiliary context C, and argues that this quantity unifies existing heuristics for clarification (Zhang and Choi; Hou et al.), prompt template selection (Sorensen et al.), and in-context demonstration selection (Ling et al.). The paper then sketches a research agenda for jointly reducing multiple uncertainty sources via a controllable proxy variable, deriving a lower bound on effective aleatoric uncertainty, identifying or rejecting uncertainty sources a priori, and using experiments to probe latent model properties.

Significance. If the central identification were valid, the paper would provide a useful unifying reframing of several ad hoc uncertainty-reduction methods and a principled justification for active clarification, retrieval, and prompt refinement rather than passive abstention. The paper is clearly written and engages honestly with limitations, explicitly acknowledging in Section 4.2 that simulators are not always available or reliable and in Section 5 the difficulties of priors and Knightian uncertainty. The information-theoretic background in Section 2.2 and the variance-correction calculation in Appendix C are correct as derivations. The research agenda, especially the lower-bound idea based on a controllable proxy variable in Section 4.1, is genuinely interesting. However, the core analogy between BME's parameter of interest and an LLM's predictive output is not established, and one supporting identity in Section 4.1 is incorrect. The paper is therefore more persuasive as an advocacy essay and literature review than as the theoretical unification it claims to provide.

major comments (3)
  1. [§3.2, Eq. (10)] The object I[Y|X;C] is conditional mutual information between the model's own output variable Y and the context C, not expected information gain about a fixed unknown state of the world, which is what BME in Section 2.2 is formulated for. Because Y is the model's random output, Eq. (10) can be positive even when C carries zero information about the correct answer. Concretely, let the true answer T be Bernoulli(1/2), let C be Bernoulli(1/2) independent of T, and let the LLM satisfy P(Y=C|X)=0.9 and P(Y≠C|X)=0.1. Then H[Y|X]=1 bit, H[Y|X,C]=H(0.9), so Eq. (10) reports roughly 0.53 bits of "reducible uncertainty" even though H[T|X,C]=H[T|X]=1 bit. The framework would recommend asking a clarification although no clarification can reduce uncertainty about T. The Section 4.2 caveat about unreliable simulators is a separate, downstream issue; even a perfect simulator P(C|X) would not turn I[Y;C|X] into an estimate of I[T;C|X]. The claimed unification of clarification, template, and demonstration heuristics in Section 3 therefore currently relabels prompt-sensitivity as epistemic reducibility. To support the central claim, the authors need to define a latent target variable T (e.g., user intent or a truth value) and either justify that P_LM(Y|X,C) approximates P(T|X,C) or explicitly restrict the claims to reducibility of model-output uncertainty.
  2. [§3.3.2, Eq. (12) and Appendix C] The reinterpretation of Sorensen et al.'s template-selection score as H[P(Y|X)] - H[P(Y|X,τ_i)] is not justified. If X denotes the original input and τ_i is a deterministic template transform, then the pair (X, τ_i(X)) generates the same information as X alone, so H[Y|X,τ_i]=H[Y|X] and the proposed difference is identically zero. The positive quantity estimated by Sorensen et al. is, on its face, a mutual information between the transformed input and the output, H[Y]-H[Y|τ_i(X)], not a reduction in conditional entropy given the original input. Moreover, Appendix C shows only that the entropy of the empirical mixture H[1/N Σ_j P(Y|x_j)] equals the conditional entropy plus a variance-correction term; it does not establish the equality to H[P(Y|X)] used in the surrounding text. This weakens the BME-unification claim for prompt templating and should be repaired or removed.
  3. [§4.1, Eq. (13)] The additivity decomposition I(Y;(A,B,E)|X)=I(Y;A|X)+I(Y;B|X)+I(Y;E|X) does not follow from the stated conditional independence of A, B, and E given X. The chain rule gives I(Y;A,B,E|X)=I(Y;A|X)+I(Y;B|X,A)+I(Y;E|X,A,B); additional assumptions such as Y⊥B|X,A and Y⊥E|X,A,B are required. The data-processing bound in Eq. (14) is correct, but Eq. (13) is a false identity in a section that proposes this decomposition as the starting point for jointly managing uncertainty sources. The equation should be corrected or explicitly labeled as an idealized extra assumption.
minor comments (4)
  1. [§3.2, Eq. (10) and §3.3.1] The notation in Eq. (10) uses E_{P(C)} but the surrounding discussion treats X as fixed; the expectation should be over P(C|X) or this should be stated explicitly. Section 3.3.1 writes H[P(Y|x,C)] without an expectation over C, which is inconsistent with Eq. (10).
  2. [§4.1] The Bayes-net diagram "A↗ ↘X→B→Ψ→Y↘ ↗E" is difficult to interpret as a standard DAG and should be redrawn with explicit nodes and edges. The sentence "We acknowledge that while this setting is oversimplistic, unlikely to occur in the real world, and do not lead to uncertainty reduction" also appears to contain a grammatical error.
  3. [§3.3.2 and Appendix C] The notation H[P(Y|X)] is used ambiguously to mean both the conditional entropy and the entropy of the marginal distribution in different places; this ambiguity is material in the discussion of Sorensen et al. and should be resolved with consistent notation.
  4. [References] The reference list for the Phan et al. entry appears to contain an extremely long author list that is likely intended for a different paper (Humanity's Last Exam); the citation should be checked and brought into line with the text.

Circularity Check

2 steps flagged · score 5.0 of 10

The claimed BME unification is largely a definitional relabeling: Eq. 10 replaces BME's true-state parameter with the LLM's own output Y, so 'reducible epistemic uncertainty' coincides by construction with I(Y;C|X), and existing heuristics are then 'unified' by retrofitting rather than by derivation.

  1. self definitional [Section 2.2 (last paragraph) and Section 3.2, Eq. 10]
    "Under the unified view of BME, epistemic uncertainty is the uncertainty reducible via experimentation. ... Using the BME perspective, we can represent the uncertainty of P_LM(Y|X) reducible with C as the experiment stimulus as I[Y|X;C] = H[P_LM[Y|X]] - E_{P(C)} H[P_LM(Y|X,C)]."

    In Section 2.2, BME is defined for gaining knowledge about a true state of the world Xi, with information gain I[Xi;Lambda]. Eq. 10 silently substitutes the LLM's own output Y for Xi and calls I[Y|X;C] the uncertainty of P_LM(Y|X) reducible with C. Because the paper has just defined epistemic uncertainty as 'uncertainty reducible via experimentation,' any gap H[P(Y|X)] - E H[P(Y|X,C)] becomes epistemic-reducible by construction. The framework therefore does not derive that this gap measures uncertainty about a true answer; it labels output-context sensitivity as epistemic reducibility, and the subsequent 'unification' inherits this definitional move.

  2. renaming known result [Section 3.3.2 (Sorensen et al. discussion)]
    "It turns out, in fact, H[1/N sum_j P(Y|x_j)] approx H[P(Y|X)] (details in Appendix C), hence their computation is also an approximation of H[P(Y|X)] - H[P(Y|X, tau_i)]. This makes the method by Sorensen et al. (2022) interpretable under the BME framework, which can explain its effectiveness."

    The paper shows that Sorensen et al.'s mutual-information template score equals an approximation of Eq. 10 and then asserts that this 'can explain its effectiveness.' The explanatory direction is only notational: a known heuristic's objective is renamed as BME information gain. No independent derivation from BME predicts template effectiveness; the formal equivalence is the entire content of the claimed 'explanation.' The same retrofitting pattern is used in Sections 3.3.1 and 3.3.3 to 'unify' clarification and in-context demonstration heuristics.

full rationale

The paper is a position piece with no fitted parameters and no empirical predictions, so the fitted-input-called-prediction pattern does not occur. The BME core is classical (Lindley 1956; DeGroot 1962), and the citation to Rainforth et al. (2024), though co-authored by one of the present authors, is not load-bearing for the mathematical framework because the framework does not depend on that survey. The main circularity is conceptual and definitional: Section 2.2 defines epistemic uncertainty as reducible-by-experimentation; Section 3.2 defines the reducible uncertainty of P_LM(Y|X) as I[Y|X;C] by taking the model output Y as the BME parameter of interest. From that point, every existing method computing H[P(Y|X)] - H[P(Y|X,C)] is automatically BME-consistent, and the paper presents this retrofitting as unification. That is a relabeling of prompt/output sensitivity as epistemic reducibility rather than a derivation that clarifying, re-templating, or adding demonstrations reduces uncertainty about a true answer. The paper is transparent about the retrofitting ('By retrofitting these examples under the BME framework'), and the individual information-theoretic identities are mathematically valid, so the circularity is partial and located in the framing rather than in a fabricated empirical result. Score 5 reflects a central claim that substantially reduces to a definitional relabeling while leaving the underlying identities intact.

Assumptions & free parameters 0 free parameters · 3 assumptions · 1 invented entities

The framework has no fitted parameters and its mathematical content is standard information theory. The key domain assumptions are the reliability of LLM self-simulation and the adequacy of entropy as an uncertainty measure; both are explicitly acknowledged as limitations. The only invented entity is an explicitly hypothetical latent variable used for an illustrative bound.

assumptions (3)
  • domain assumption The LLM's predictive distribution P_LM(Y|X,C) is a faithful model of the true conditional distribution of outcomes given context.
    BME requires a specified experiment outcome distribution; the paper uses the LLM itself to simulate P(C|X) and P(Y|X,C), assuming these are reliable. Discussed in Sections 3.2 and 4.2.
  • domain assumption Shannon entropy is an appropriate measure of uncertainty for LLM outputs.
    The framework quantifies uncertainty via entropy (Eq. 1) without justifying its adequacy for sequence- or semantic-level outputs. Section 2.2.
  • standard math Standard information-theoretic identities (mutual information, data processing inequality, KL divergence) hold for the random variables considered.
    Used throughout, e.g., Eq. 9, Eq. 13, Eq. 14. These are standard.
invented entities (1)
  • Controllable proxy variable Ψ
    purpose: Hypothetical latent variable in the Bayes net of Section 4.1 that combines effects of ambiguity, prompt templates, and in-context examples to derive an upper bound on reducible uncertainty.
    The paper itself says the setting is 'oversimplistic, unlikely to occur in the real world', so Ψ is an illustrative construct, not a claimed real entity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Extending Epistemic Uncertainty Beyond Parameters Would Assist in Designing Reliable LLMs." pith.science (2026). https://pith.science/paper/2DRBPSBJ

@misc{pith2026250607448,
  author       = {Pith},
  title        = {Pith review of: Extending Epistemic Uncertainty Beyond Parameters Would Assist in Designing Reliable LLMs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2DRBPSBJ}},
  note         = {Machine review of arXiv:2506.07448}
}
read the original abstract

Although large language models (LLMs) are highly interactive and extendable, current approaches to ensure reliability in deployments remain mostly limited to rejecting outputs with high uncertainty in order to avoid misinformation. This conservative strategy reflects the current lack of tools to systematically distinguish and respond to different sources of uncertainty. In this paper, we advocate for the adoption of Bayesian Modeling of Experiments -- a framework that provides a coherent foundation to reason about uncertainty and clarify the reducibility of uncertainty -- for managing and proactively addressing uncertainty that arises in LLM deployments. This framework enables LLMs and their users to take contextually appropriate steps, such as requesting clarification, retrieving external information, or refining inputs. By supporting active resolution rather than passive avoidance, it opens the door to more reliable, transparent, and broadly applicable LLM systems, particularly in high-stakes, real-world settings.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

38 extracted references · 14 canonical work pages

  1. [4]

    ArXiv:2403.04132 [cs]

    Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference. ArXiv:2403.04132 [cs]. URLhttp://arxiv.org/abs/2403.041322, 10 10 Cover T, Thomas J. 2012.Elements of Information Theory. Wiley. ISBN 978-1-118-58577-1. 4, 8 DeGroot MH

    arXiv 2012

  2. [6]

    InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)

    Context versus Prior Knowledge in Language Models. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Bangkok, Thailand: Association for Computational Linguistics, 13211–13235. URLhttps://aclanthology.org/2024.acl-long.7148, 9 Farquhar S, Kossen J, Kuhn L, Gal Y

    work page 2024

  3. [7]

    URLhttps://arxiv.org/abs/2501.142492, 10 Rainforth T, Foster A, Ivanova DR, Bickford Smith F

    arXiv

  4. [9]

    ArXiv:1112.5745 [cs, stat]

    Bayesian Active Learning for Classification and Preference Learning. ArXiv:1112.5745 [cs, stat]. URLhttp://arxiv.org/abs/1112.574518 Kadavath S, Conerly T, Askell A, Henighan T, Drain D, Perez E, Schiefer N, Hatfield-Dodds Z, DasSarma N, Tran-Johnson E, Johnston S, El-Showk S, Jones A, Elhage N, Hume T, Chen A, Bai Y , Bowman S, Fort S, Ganguli D, Hernand...

    arXiv

  5. [11]

    InProceedings of the 56th Annual ACM Symposium on Theory of Computing

    Calibrated Language Models Must Hallucinate. InProceedings of the 56th Annual ACM Symposium on Theory of Computing. Vancouver BC Canada: ACM. ISBN 979-8-4007-0383-6, 160–171. URLhttps://dl.acm.org/doi/10.1145/3618260.36497779 11 Kambhampati S

    arXiv

  6. [12]

    ACM64: 31–32

    Polanyi’s revenge and AI’s new romance with tacit knowledge.Commun. ACM64: 31–32. ISSN 0001-0782. URLhttps://dl.acm.org/doi/10.1145/344636910 Kapoor S, Gruver N, Roberts M, Collins K, Pal A, Bhatt U, Weller A, Dooley S, Goldblum M, Wilson AG

    work page doi:10.1145/344636910

  7. [13]

    ArXiv:2406.08391 [cs, stat]

    Large Language Models Must Be Taught to Know What They Don’t Know. ArXiv:2406.08391 [cs, stat]. URLhttp://arxiv.org/abs/2406.083915 Kendall A, Gal Y

    arXiv

  8. [15]

    ArXiv:2406.15927 [cs]

    Semantic Entropy Probes: Robust and Cheap Hallucination Detection in LLMs. ArXiv:2406.15927 [cs]. URLhttp://arxiv.org/abs/2406.159275 Krug S. 2013.Don’t make me think, revisited: a common sense approach to web usability. V oices that matter. Pearson Education. ISBN 978-0-13-359726-4. Tex.lccn: 2014397947. URLhttps://books.google.com/books?id=QlduAgAAQBAJ2...

    arXiv 2013

Show all 38 references
  1. [17]

    URLhttps://openreview.net/forum?id=8s8K2UZGTZ1, 4 Lin Z, Trivedi S, Sun J

    Teaching Models to Express Their Uncertainty in Words.Transac- tions on Machine Learning ResearchISSN 2835-8856. URLhttps://openreview.net/forum?id=8s8K2UZGTZ1, 4 Lin Z, Trivedi S, Sun J. 2024a. Contextualized Sequence Likelihood: Enhanced Confidence Scores for Natural Languag...

    arXiv

  2. [19]

    Uncertainty Quantification for In-Context Learning of Large Language Models. In Duh K, Gomez H, Bethard S (eds.)Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers...

    2024

  3. [20]

    ArXiv:2504.01990 [cs]

    Advances and Challenges in Foundation Agents: From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systems. ArXiv:2504.01990 [cs]. URLhttp://arxiv.org/abs/2504.019902 Lu C, Lu C, Lange RT, Foerster J, Clune J, Ha D

    arXiv

  4. [21]

    ArXiv:2408.06292 [cs]

    The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery. ArXiv:2408.06292 [cs]. URLhttp://arxiv.org/abs/2408.062922 Lu Y , Bartolo M, Moore A, Riedel S, Stenetorp P

    arXiv

  5. [22]

    ArXiv:2104.08786 [cs]

    Fantastically Ordered Prompts and Where to Find Them: Overcoming Few-Shot Prompt Order Sensitivity. ArXiv:2104.08786 [cs]. URLhttp://arxiv.org/abs/2104.087866 Malinin A, Gales M

    arXiv

  6. [23]

    In International Conference on Learning Representations

    Uncertainty Estimation in Autoregressive Structured Prediction. In International Conference on Learning Representations. URLhttps://openreview.net/forum?id=jN5y-zb5Q7m4 Montgomery DC. 2017.Design and Analysis of Experiments. John Wiley & Sons. ISBN 978-1-119- 11347-8. Google-B...

    2017

  7. [24]

    ISSN 2377-3766

    CLARA: Classifying and Disambiguating User Commands for Reliable Interactive Robotic Agents.IEEE Robotics and Automation Letters 9: 1059–1066. ISSN 2377-3766. URLhttps://ieeexplore.ieee.org/document/103369012, 5, 6 Phan L, Gatti A, Han Z, Li N, Hu J, Zhang H, Zhang CBC, Shaaba...

  8. [26]

    Statistical Science39

    Modern Bayesian Experimental Design. Statistical Science39. ISSN 0883-4237. URL https://projecteuclid.org/journals/statistical-science/volume-39/issu e-1/Modern-Bayesian-Experimental-Design/10.1214/23-STS915.full2, 4, 18 Smith FB, Kossen J, Trollope E, Wilk Mvd, Foster A, Rainforth T

    doi

  9. [27]

    ArXiv:2412.20892 [cs]

    Rethinking Aleatoric and Epistemic Uncertainty. ArXiv:2412.20892 [cs]. URLhttp://arxiv.org/abs/2412.2089218, 19 15 Sorensen T, Robinson J, Rytting C, Shaw A, Rogers K, Delorey A, Khalil M, Fulda N, Wingate D

    arXiv

  10. [28]

    InProceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)

    An Information-theoretic Approach to Prompt Engineering Without Ground Truth Labels. InProceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Dublin, Ireland: Association for Computational Linguistics, 819–862. URLhttps...

    2022

  11. [30]

    URL https://proceedings.neurips.cc/paper/2017/hash/2650d6089a6d640c5e85b 2b88265dc2b-Abstract.html18, 19 Kiureghian AD, Ditlevsen O

    Curran Associates, Inc. URL https://proceedings.neurips.cc/paper/2017/hash/2650d6089a6d640c5e85b 2b88265dc2b-Abstract.html18, 19 Kiureghian AD, Ditlevsen O

    2017

  12. [31]

    InProceedings of the 2023 Conference on Empirical Methods in Natural Language Processing

    Just Ask for Calibration: Strategies for Eliciting Calibrated Confidence Scores from Language Models Fine- Tuned with Human Feedback. InProceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. Singapore: Association for Computational Linguistics,...

    2023

  13. [32]

    ArXiv:2502.13069 [cs]

    Interactive Agents to Overcome Ambiguity in Software Engineering. ArXiv:2502.13069 [cs]. URLhttp://arxiv.org/abs/2502.130692 Wang Z, Holmes C

  14. [33]

    URL https://proceedings.neurips.cc/paper/2020/hash/1457c0d6bfcb4967418bf b8ac142f64a-Abstract.html1, 6 Carlini N

    Curran Associates, Inc., 1877–1901. URL https://proceedings.neurips.cc/paper/2020/hash/1457c0d6bfcb4967418bf b8ac142f64a-Abstract.html1, 6 Carlini N

    1901

  15. [34]

    ArXiv:2406.05213 [cs, stat]

    On Subjective Uncertainty Quantification and Calibration in Natural Language Generation. ArXiv:2406.05213 [cs, stat]. URLhttp://arxiv.org/abs/2406.0521319 Wei J, Tay Y , Bommasani R, Raffel C, Zoph B, Borgeaud S, Yogatama D, Bosma M, Zhou D, Metzler D, Chi EH, Hashimoto T, Vin...

    arXiv

  16. [35]

    InThe Twelfth International Conference on Learning Representations

    Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs. InThe Twelfth International Conference on Learning Representations. URLhttps://openreview.net/forum?id=gjeQKFxFpZ1, 4 Yadkori Y A, Kuzborskij I, György A, Szepesvári C. 2024a. To Bel...

    2024

  17. [36]

    InProceedings of the 2023 CHI Conference on Human Factors in Computing Systems

    Why Johnny Can’t Prompt: How Non- AI Experts Try (and Fail) to Design LLM Prompts. InProceedings of the 2023 CHI Conference on Human Factors in Computing Systems. Hamburg Germany: ACM. ISBN 978-1-4503-9421-5, 1–21. URLhttps://dl.acm.org/doi/10.1145/3544548.35813882 Zhang MJQ, Choi E

    2023

  18. [37]

    58077–58117. URL https://papers.nips.cc/paper_files/paper/2024/hash/6aebba00fff5b6de7 b488e496f80edd7-Abstract-Conference.html8, 9 Yadkori Y A, Kuzborskij I, Stutz D, György A, Fisch A, Doucet A, Beloshapka I, Weng WH, Yang YY , Szepesvári C, Cemgil AT, Tomasev N. 2024b. Mitig...

    2024 arXiv

  19. [38]

    ArXiv:2311.09469

    Clarify When Necessary: Resolving Ambiguity Through Interaction with LMs. ArXiv:2311.09469. URLhttp://arxiv.org/abs/2311.094692, 5, 6 Zhao TZ, Wallace E, Feng S, Klein D, Singh S

    arXiv

  20. [39]

    ArXiv:2102.09690 [cs]

    Calibrate Before Use: Improving Few-Shot Performance of Language Models. ArXiv:2102.09690 [cs]. URLhttp://arxiv.org/abs/2102.096906 17 APPENDIX A What Is Reducible Depends on the Experiment Apparatus In this section, we show thatthe BME framework is compatible with classical d...

    2024 arXiv

  21. [40]

    tuning knob

    more specifically distinguishes compo- nents of an experimentΛ into its design X and outcome (unknown before performing the experiment) Y , i.e., an experiment λ is representable as a tuple (x, y). It is then customary to assume Y and Ξ together specifyP(Y|Ξ), whileΞ⊥XandP(Ξ|X...

    2011

  22. [1956]

    ISSN 0003-4851, 2168-8990

    On a Measure of the Information Provided by an Experiment.The Annals of Mathematical Statistics27: 986–1005. ISSN 0003-4851, 2168-8990. Publisher: Institute of Mathematical Statistics. URL https://projecteuclid.org/journals/annals-of-mathematical-statistics/ volume-27/issue-4/...

  23. [1962]

    ISSN 0003-4851, 2168-8990

    Uncertainty, Information, and Sequential Experiments.The Annals of Mathe- matical Statistics33: 404–419. ISSN 0003-4851, 2168-8990. Publisher: Institute of Mathematical Statistics. URL https://projecteuclid.org/journals/annals-of-mathematical-statistics/ volume-33/issue-2/Unce...

  24. [2011]

    InPerspectives on Thinking, Judging and Decision Making: A Tribute to Karl Halvor Teigen

    Distinguishing Two Dimensions of Uncertainty. InPerspectives on Thinking, Judging and Decision Making: A Tribute to Karl Halvor Teigen. Universitetsforlaget. ISBN 978-82-15-01878-2. Google-Books-ID: 9NDiygAACAAJ. 3 Gal Y . 2016.Uncertainty in Deep Learning. Ph.D. thesis, Unive...

    2016

  25. [2016]

    InProceedings of The 33rd International Conference on Machine Learning

    Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning. InProceedings of The 33rd International Conference on Machine Learning. PMLR, 1050–1059. ISSN: 1938-7228. URLhttps://proceedings.mlr.press/v48/gal16.html5 Gupta N, Narasimhan H, Jitkrittum W,...

    1938

  26. [2021]

    URLhttps://arxiv.org/abs/2107.03374v21 Chiang WL, Zheng L, Sheng Y , Angelopoulos AN, Li T, Li D, Zhang H, Zhu B, Jordan M, Gonzalez JE, Stoica I

    Evaluating Large Language Models Trained on Code. URLhttps://arxiv.org/abs/2107.03374v21 Chiang WL, Zheng L, Sheng Y , Angelopoulos AN, Li T, Li D, Zhang H, Zhu B, Jordan M, Gonzalez JE, Stoica I

    arXiv

  27. [2022]

    ArXiv:2207.05221 [cs]

    Language Models (Mostly) Know What They Know. ArXiv:2207.05221 [cs]. URLhttp://arxiv.org/abs/2207.052211, 4, 5 Kalai AT, Vempala SS

    arXiv

  28. [2023]

    ArXiv:2206.04615 [cs]

    Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models. ArXiv:2206.04615 [cs]. 16 URLhttp://arxiv.org/abs/2206.046152, 10 Tian K, Mitchell E, Zhou A, Sharma A, Rafailov R, Yao H, Finn C, Manning C

    arXiv

  29. [2024]

    ArXiv:2406.04306 [cs]

    Semantically Diverse Language Generation for Uncertainty Estimation in Language Models. ArXiv:2406.04306 [cs]. URLhttp://arxiv.org/abs/2406.043065 Berend D, Kontorovich A

  30. [2025]

    ArXiv:2501.13075 [cs]

    Evolution and The Knightian Blindspot of Machine Learning. ArXiv:2501.13075 [cs]. URLhttp://arxiv.org/abs/2501.130752, 9 Li R, Allal LB, Zi Y , Muennighoff N, Kocetkov D, Mou C, Marone M, Akiki C, Li J, Chim J, Liu Q, Zheltonozhskii E, Zhuo TY , Wang T, Dehaene O, Lamy-Poirier...

    arXiv

Pith tools

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