Pith. sign in

REVIEW 4 major objections 6 minor 2 cited by

Prune 'n Predict: Optimizing LLM Decision-making with Conformal Prediction

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read The paper claims that pruning an MCQ's answer choices to a conformal prediction set and re-asking the LLM improves accuracy, with learned scores (CP-OPT) making the gains larger.

desk verdict A useful, clearly-written test-time pruning idea whose main empirical claim is undercut by a re-labeling confound and an overbroad abstract. read the letter →

arxiv 2501.00555 v2 pith:T7KL6SWS submitted 2024-12-31 cs.LG cs.AIstat.APstat.ML

Harit Vishwakarma , Alan Mishler , Thomas Cook , Niccolò Dalmasso , Natraj Raman , Sumitra Ganesh This is my paper · ORCID
classification cs.LGcs.AIstat.APstat.ML
keywords conformalpredictionlargelanguagemodelsmultiple-choicequestionansweringsetsscoreoptimizationaccuracyimprovementuncertaintyquantificationrevision
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

The paper tries to establish a test-time method for improving LLM accuracy on multiple-choice tasks without retraining: use conformal prediction to build a small set of answer choices that contains the correct answer with high probability, then re-ask the LLM with only those choices. It calls this conformal revision of questions (CROQ) and claims it improves accuracy over standard inference in most settings. Because raw logit scores often produce large sets, the paper also introduces CP-OPT, a post-hoc optimization that learns a score function to shrink prediction sets while keeping the conformal coverage guarantee. The empirical claim is that CROQ works, and works better with CP-OPT than with logit scores, on MMLU, ToolAlpaca, and TruthfulQA across three small instruction-tuned LLMs.

What carries the argument

The central object is the conformal prediction set C(x; g, tau) = {y : g(x, y) >= tau}, used to prune the MCQ. The argument runs through the identity $\Delta$ = sum_{k=1}^{m} r_k rho_k f_post(k) - a, where a is baseline accuracy, r_k is the share of questions whose prediction set has size k, rho_k is the coverage among those questions, and f_post(k) is the LLM's accuracy on a k-option revised question. Proposition 3.2 shows that under Assumption 3.1, which posits that f_post(k) is monotonically decreasing in k, the accuracy gain is maximized by concentrating r_k rho_k on small k, and the proof reduces this to a fractional knapsack problem. CP-OPT is the engine that pursues that concentration: it replaces hard set-size indicators with sigmoid surrogates and solves a differentiable penalty problem over a small neural network that takes the LLM's last-token logits and penultimate-layer representation as input.

What would settle it

Measure the conditional accuracy f_post(k) on a held-out MCQ benchmark with enough items per set size and compare overall CROQ accuracy with single-round inference on the same items. If f_post(k) does not decrease with k (for example, if revised 2-option questions are less accurate than revised 3-option questions) or if the paired difference is not positive, then Assumption 3.1 is violated for that setting and the paper's theoretical reason for CROQ helping disappears.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that an LLM answering a multiple-choice question can be made more accurate by a two-step procedure: build a conformal prediction set over answer choices that covers the true answer with probability at least 1 - alpha, then rewrite the question using only the choices in that set and prompt the LLM again. The paper shows that this procedure, CROQ, improves accuracy over single-round inference in 19 of 27 settings with logit scores and in 24 of 27 settings with CP-OPT scores, with statistically significant gains concentrated on questions with 10 or 15 options. The paper further claims that CP-OPT, a neural-network score function trained to minimize expected set size subject to coverage, produces smaller conformal sets than raw logits in most settings while keeping coverage near the 95% target, and that CROQ with CP-OPT outperforms CROQ with logits in 22 of 27 settings. The argument is carried by the decomposition of post-CROQ accuracy into a sum over prediction-set sizes, showing that gains come from shifting probability mass toward smaller sets.

Load-bearing premise

The argument assumes that an LLM's accuracy on a revised question declines as more answer choices remain, so that pruning choices reliably makes the task easier; if re-asking with fewer options does not actually improve accuracy, the central mechanism for CROQ's gains collapses.

Editorial extensions

If this is right

  • CROQ can be applied to any pretrained LLM with no fine-tuning, so deployment can improve accuracy by adding a conformal scoring step and a second prompt.
  • Because the conformal coverage guarantee is distribution-free, users can choose alpha to trade a bounded chance of deleting the correct answer against the accuracy gain from smaller sets.
  • Using CP-OPT scores instead of raw logits reduces average conformal set sizes in most of the 27 settings, which makes the downstream revision more effective.
  • In the NL2SQL agentic workflow, the same conformal pruning reduced LLM cost by about 45 percent at comparable or better accuracy, so cost and accuracy gains can come together.
  • Accuracy gains are uneven: they are larger on 10- and 15-option questions and smaller or slightly negative on 4-option questions, so practitioners should tune alpha and verify gains per task.

Reading between the lines

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

  • The paper's own conditional-accuracy tables show that f_post(k) is not strictly monotone for some models and datasets, so a more general version of the argument might need only a weighted-average or stochastic monotonicity; that weakening is not proved in the paper.
  • Because CP-OPT consumes only last-token logits and hidden states, it could be inserted into serving stacks as a lightweight pre-answering filter; the paper reports accuracy and set sizes, not serving latency, so that cost remains untested.
  • The same prune-and-reask loop could be applied to tool or API selection with very large candidate sets, and the NL2SQL experiment is an early indication; whether gains persist when there are hundreds of candidate actions is not tested.
  • A natural stress test would be to run CROQ with a score function that is deliberately uninformative, such as uniform random scores, to confirm that the accuracy gain is driven by set-size reduction rather than by hidden information in the conformal score.
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

4 major / 6 minor

Summary. The paper proposes CROQ (conformal revision of questions), a two-stage procedure for multiple-choice LLM tasks: first, split conformal prediction is used to build a prediction set of answer options; second, the question is revised to contain only the options in that set, the surviving options are re-keyed to consecutive letters starting at A, and the LLM is prompted again. The authors argue that fewer options improve accuracy, and they introduce CP-OPT, a post-hoc score-learning framework that minimizes expected conformal set size while preserving marginal coverage. Experiments on MMLU, ToolAlpaca, and TruthfulQA with three instruction-tuned LLMs and 4/10/15 options compare accuracy before/after CROQ using logit scores and CP-OPT scores, and report set-size and coverage comparisons. The central claims are that CROQ improves accuracy over standard inference, that CP-OPT reduces set sizes at comparable coverage, and that CROQ with CP-OPT outperforms CROQ with logits.

Significance. If the causal effect were cleanly established, CROQ would be a practically valuable, test-time accuracy booster with a formal coverage guarantee, and CP-OPT would be a useful contribution to conformal score design for LLMs. The paper has real strengths: Proposition 2.1 gives a standard split-conformal coverage proof; the experiments use paired tests and report per-setting conditional tables; the NL2SQL application in Appendix C.3 demonstrates a cost reduction with preserved accuracy; and the set-size histograms give a clear picture of where CP-OPT changes the distribution of conformal sets. However, the central causal claim is currently confounded: the 'after' condition differs from the baseline in three ways at once (pruning, re-keying of options, and a second forward pass), so the reported gains cannot be attributed to pruning alone. The manuscript's own conditional data also undermine the monotonicity assumption that motivates the theoretical analysis. The contribution is promising and the issues are addressable, but the paper as written overstates its central claim.

major comments (4)
  1. [§3.1, Step 2; Tables 2 and 5] The CROQ 'after' condition changes three things relative to the baseline: the option set is pruned, the surviving options are re-keyed to contiguous letters starting at A, and a second forward pass is used. Since the LLM answer is the greedy argmax over option keys (Eq. 12), and the paper itself cites Zheng et al. (2024) on LLMs' sensitivity to option ordering and key labeling, the accuracy gains in Tables 2 and 5 cannot be attributed to pruning alone. Please add control conditions: (i) re-ask the original unpruned question as a second-pass control; (ii) prune the option set but preserve the original keys (e.g., keep {C,D} rather than re-keying to {A,B}); and (iii) ideally, also vary the order of the surviving options. At minimum, report how much of the gain survives when re-keying is removed. Without such controls, the central claim that pruning choices improves accuracy is not established.
  2. [§3.1, Assumption 3.1; Table 6] Assumption 3.1 (monotone conditional accuracy) is load-bearing for Proposition 3.2, but the manuscript's own conditional tables do not support strict monotonicity. In Table 6 (Gemma-2, MMLU-15, Ours), the conditional accuracy on questions where the correct answer is retained, computed as Acc. After / Coverage, is about 62.2% at k=8 (56.14/90.23) and about 62.2% at k=9 (57.49/92.40), and adjacent sizes show non-monotone dips (e.g., k=7 at 60.26/89.74, about 67.1%, is lower than k=6 at 70.81/90.54, about 78.2%). Moreover, the tables report raw 'Acc. After', which includes questions where the correct answer was pruned, rather than the conditional quantity fpost(k) defined in the text; fpost(k) is never estimated or reported. Please either test Assumption 3.1 empirically or explicitly relegate Proposition 3.2 to a conditional statement under an idealization, and report conditional accuracy separately from coverage.
  3. [Abstract; Contribution 1; §4.2, H2] The abstract and Contribution 1 state that CROQ 'improves accuracy over the standard inference' and 'consistently improves accuracy', but Tables 2 and 5 show zero or negative gains in 8/27 and 3/27 settings respectively, including all three standard 4-option MMLU settings with logits (gains of -0.19, -1.19, and +0.07). The paper's own discussion in §4.2 correctly says 'in the vast majority of settings' and 'rare small drops'; the front matter should be aligned with those hedged claims. In addition, please report an aggregate summary across the 27 settings (e.g., mean gain with a confidence interval or a sign-test result) so that the reader can assess the overall effect rather than relying on the count of individually significant cells.
  4. [H1, Table 1; Contribution 2] Contribution 2 claims CP-OPT yields smaller average set sizes 'at the same level (95%) of coverage', but Table 1 shows that in 6/27 settings the CP-OPT set-size reduction is accompanied by a statistically significant decrease in coverage relative to logits (e.g., MMLU-4 Gemma-2: 94.23 vs 95.16; MMLU-10 Gemma-2: 94.04 vs 95.00; TruthfulQA-4 Llama-3: 92.41 vs 95.95). The split-conformal guarantee is not invalidated because the threshold is recalibrated on a separate set, but the empirical comparison is not 'same coverage' in the settings where coverage drops significantly. Please either compare at a common empirical coverage level, or explicitly state the trade-off and reword the contribution to say 'comparable or slightly lower marginal coverage, with the differences listed in Table 1.'
minor comments (6)
  1. [After Eq. (9)] The text says 'by the strong law of larger numbers'; this should be 'strong law of large numbers'.
  2. [§3.2, definition of G] The architecture description uses d3 = (d + m)/4 in the text but the layer dimensions are W1 ∈ R^{d0×d1}, W2 ∈ R^{d1×d2}, W3 ∈ R^{d2×m}; the intended hidden dimension is d2 = (d + m)/4, so the notation is inconsistent and should be fixed.
  3. [§3.2, Eq. (P2)] The regularization term λ1||g||2^2 is introduced without a definition of λ1, and Table 21 lists 'weight decay' rather than λ1; please clarify whether weight decay is λ1 and report its values in the hyperparameter table.
  4. [Proposition 3.2, item 2] The sufficient condition is written as 'rkρk > a/(mfpost(k))', which is ambiguous; it should be typeset as rkρk > a / (m · fpost(k)) to match the proof in Appendix B.2.
  5. [Appendix C.2] The MMLU-Pro result is reported only as a '3% relative improvement' over a 36.4% baseline; please report the actual after-CROQ accuracy and the number of test questions so the reader can judge the magnitude.
  6. [Appendix D] The significance analysis uses per-setting paired t-tests at level 0.05 but does not adjust for multiple comparisons across the 27 settings; please note this or apply a simple correction (e.g., Benjamini-Hochberg) and state whether the conclusions change.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: accuracy gains are empirically measured, the conformal coverage proof is self-contained, and Proposition 3.2 is an algebraic identity under an explicit assumption, not a fitted prediction.

full rationale

The paper's central claims are empirical comparisons on held-out test splits. CROQ accuracy after revision is measured, not derived from a fitted parameter: CP-OPT optimizes a surrogate objective for set-size reduction on a training split, but the accuracy gains in Tables 2, 3, and 5 are computed from test-set measurements with a separately calibrated conformal threshold (Appendix D). Proposition 3.2 is a law-of-total-probability decomposition followed by a knapsack argument; it does not define the gain as equal to set-size reduction, and it is explicitly conditional on Assumption 3.1 (Monotone Accuracy). The paper itself acknowledges this assumption is a point of departure for future analysis, and its own conditional tables, e.g., Table 6, show non-monotonic empirical fpost values, which weakens the theory but is not circularity. The conformal coverage proof in Appendix B.1 is a standard exchangeability argument and is self-contained. Self-citations in related work and Appendix C.4 are not load-bearing; no uniqueness theorem or ansatz is imported from the authors' prior work. The re-labeling of surviving option keys to start at 'A' and the second forward pass are possible confounds for isolating the pruning effect, but they are not circular reductions of the claimed accuracy gain to the paper's inputs.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

CP-OPT's network weights are fitted to data, and per-setting hyperparameters (lambda, lr, weight decay) are hand-tuned; these are the main fitted quantities. The conformal threshold is a data-dependent quantile and not a free parameter. The scientific claims rest on exchangeability and on a monotonicity assumption that the paper states but does not establish.

free parameters (4)
  • CP-OPT penalty weight lambda = 0.1 to 10.0 (varies per model, dataset, and number of options)
    Chosen by hand per setting (Table 21); controls the trade-off between surrogate set size and coverage in the nonconvex objective.
  • CP-OPT learning rate = 5e-6 to 1e-4
    Tuned per setting; affects training of the score network.
  • CP-OPT weight decay = 1e-10 to 1e-6
    Tuned per setting for L2 regularization.
  • Sigmoid temperature beta = 1.0
    Fixed for all experiments; the surrogate converges to the indicator only as beta goes to infinity, so beta=1 may bias the optimum.
assumptions (4)
  • standard math Split conformal coverage guarantee requires exchangeability of calibration and test points
    Used in Proposition 2.1; the custom resampled splits are exchangeable by construction.
  • domain assumption Assumption 3.1: fpost(k) is monotonically decreasing in k
    Used to motivate why smaller conformal sets should improve CROQ; contradicted by results in Table 6.
  • domain assumption The accuracy improvement from random distractor elimination (Figure 1) carries over to conformal subsets
    Conformal subsets are not random subsets; they consist of the most plausible options, so the remaining distractor set may be harder. The paper does not validate this transfer.
  • standard math The sigmoid surrogate with beta=1 is a valid approximation for the indicator in the optimization problem
    Convergence requires beta going to infinity; beta=1 is a finite approximation that may yield different optima.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Prune 'n Predict: Optimizing LLM Decision-making with Conformal Prediction." pith.science (2026). https://pith.science/paper/T7KL6SWS

@misc{pith2026250100555,
  author       = {Pith},
  title        = {Pith review of: Prune 'n Predict: Optimizing LLM Decision-making with Conformal Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T7KL6SWS}},
  note         = {Machine review of arXiv:2501.00555}
}
read the original abstract

Large language models (LLMs) are empowering decision-making in several applications, including tool or API usage and answering multiple-choice questions (MCQs). However, incorrect outputs pose significant risks in high-stakes domains like healthcare and finance. To quantify LLM uncertainty and thereby mitigate these risks, recent works employ conformal prediction (CP), a model- and distribution-agnostic framework that uses LLM outputs to generate a \emph{prediction set} containing the true answer with high probability. Leveraging CP, we propose \emph{conformal revision of questions} (CROQ), which revises the question by narrowing down the available choices to those in the prediction set and asking the LLM the revised question. We expect LLMs to be more accurate on revised questions with fewer choices. Furthermore, we expect CROQ to be effective when the prediction sets from CP are small. Commonly used logit scores often lead to large sets, diminishing CROQ's effectiveness. To overcome this, we propose CP-OPT, an optimization framework to learn scores that minimize set sizes while maintaining coverage. Our extensive experiments on MMLU, ToolAlpaca, and TruthfulQA datasets with multiple LLMs show that CROQ improves accuracy over the standard inference, with more pronounced gains when paired with CP-OPT.

Figures

Figures reproduced from arXiv: 2501.00555 by the authors.

Figure 1
Figure 1. Accuracy for three LLMs on the TruthfulQA [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. (CROQ) Illustration of conformal revision of questions and prompting the LLM with the revised question. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Proportion of questions deferred to a human when conformal prediction set sizes exceed a certain cutoff [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Accuracy on revised questions on the MMLU and ToolAlpaca datasets while varying miscoverage parameter [PITH_FULL_IMAGE:figures/full_fig_p018_4.png]
Figure 5
Figure 5. Figure 5: Accuracy on revised questions on the MMLU and ToolAlpaca datasets while varying miscoverage parameter [PITH_FULL_IMAGE:figures/full_fig_p019_5.png]
Figure 6
Figure 6. Figure 6: Distributions of sizes of sets obtained from CP-OPT and logit scores on MMLU dataset and Gemma-2 model. [PITH_FULL_IMAGE:figures/full_fig_p019_6.png]
Figure 7
Figure 7. Figure 7: Distributions of sizes of sets obtained from CP-OPT and logit scores on MMLU dataset and Llama-3 model. [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: Distributions of sizes of sets obtained from CP-OPT and logit scores on MMLU dataset and Phi-3 model [PITH_FULL_IMAGE:figures/full_fig_p020_8.png]
Figure 9
Figure 9. Figure 9: Distributions of sizes of sets obtained from CP-OPT and logit scores on ToolAlpaca dataset and Gemma-2 [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]
Figure 10
Figure 10. Figure 10: Distributions of sizes of sets obtained from CP-OPT and logit scores on ToolAlpaca dataset and Llama-3 [PITH_FULL_IMAGE:figures/full_fig_p020_10.png]
Figure 11
Figure 11. Figure 11: Distributions of sizes of sets obtained from CP-OPT and logit scores on ToolAlpaca dataset and Phi-3 model. [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]
Figure 12
Figure 12. Figure 12: Distributions of sizes of sets obtained from CP-OPT and logit scores on Truthful QA dataset and Gemma-2. [PITH_FULL_IMAGE:figures/full_fig_p021_12.png]
Figure 13
Figure 13. Figure 13: Distributions of sizes of sets obtained from CP-OPT and logit scores on Truthful QA dataset and Phi-3 [PITH_FULL_IMAGE:figures/full_fig_p021_13.png]
Figure 14
Figure 14. Figure 14: Distributions of sizes of sets obtained from CP-OPT and logit scores on Truthful QA dataset and Llama-3 [PITH_FULL_IMAGE:figures/full_fig_p021_14.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Domain-Shift-Aware Conformal Prediction for Large Language Models

    stat.ML 2025-10 reject novelty 5.0 of 10

    DS-CP reweights calibration scores via embedding-based density ratios to improve conformal coverage under domain shift, but its stated guarantee requires a weight condition that the default configuration violates.

  2. Smarter Together: Combining Large Language Models and Small Models for Physiological Signals Visual Inspection

    cs.AI 2025-01 conditional novelty 5.0 of 10

    ConMIL uses per-class interpretable multiple instance learning and conformal prediction to feed LLMs calibrated, highlighted hints, lifting accuracy on ECG and EEG visual inspection from 13 to 48 percent to 71 to 97 percent.

Reference graph

Works this paper leans on

27 extracted references · 10 canonical work pages · cited by 2 Pith papers

  1. [1]

    Abdin, S

    M. Abdin, S. A. Jacobs, A. A. Awan, J. Aneja, A. Awadallah, H. Awadalla, N. Bach, A. Bahree, A. Bakhtiari, H. Behl, et al. Phi-3 technical report: A highly capable language model locally on your phone. arXiv preprint arXiv:2404.14219,

    arXiv

  2. [3]

    As expected, lower set size cutoffs result in higher accuracy. As the set size cutoff increases, the accuracy approaches the LLM’s marginal accuracy, while the number of deferrals (i.e., the cost of obtaining the answer from a human or more expensive model) decreases. In the top row of the figure, the differences in the set sizes between logit and CP-OPT ...

    work page 2022

  3. [4]

    J. J. Cherian, I. Gibbs, and E. J. Cand`es. Large language model validity via enhanced conformal prediction methods. arXiv preprint arXiv:2406.09714,

    arXiv

  4. [5]

    Clarkson, W

    J. Clarkson, W. Xu, M. Cucuringu, and G. Reinert. Split conformal prediction under data contamination. arXiv preprint arXiv:2407.07700,

    arXiv

  5. [6]

    Dubey, A

    A. Dubey, A. Jauhri, A. Pandey, A. Kadian, A. Al-Dahle, A. Letman, A. Mathur, A. Schelten, A. Yang, A. Fan, et al. The llama 3 herd of models. arXiv preprint arXiv:2407.21783,

    arXiv

  6. [7]

    ISSN 2835-8856. T. Groot and M. Valdenegro Toro. Overconfidence is key: Verbalized uncertainty evaluation in large language and vision-language models. In Proceedings of the 4th Workshop on Trustworthy Natural Language Processing (TrustNLP 2024). Association for Computational Linguistics,

    work page 2024

  7. [8]

    Gupta, H

    N. Gupta, H. Narasimhan, W. Jitkrittum, A. S. Rawat, A. K. Menon, and S. Kumar. Language model cascades: Token-level uncertainty and beyond. arXiv preprint arXiv:2404.10136,

    arXiv

  8. [10]

    Krause, W

    L. Krause, W. Tufa, S. Baez Santamaria, A. Daza, U. Khurana, and P. V ossen. Confidently wrong: Exploring the calibration and expression of (un)certainty of large language models in a multilingual setting. In Proceedings of the Workshop on Multimodal, Multilingual Natural Language Generation and Multilingual WebNLG Challenge (MM-NLG 2023). Association for...

    work page 2023

Show all 27 references
  1. [11]

    After (CROQ + Logits) Acc

    18 Prune ’n Predict: Optimizing LLM Decision-making with Conformal Prediction 0.0 0.2 0.4 63 64 65Accuracy (%) MMLU-4, Llama-3 0.0 0.2 0.4 54 56 58 MMLU-10, Llama-3 0.0 0.2 0.4 52 54 56 MMLU-15, Llama-3 0.0 0.2 0.4 Coverage parameter ( ) 90 92Accuracy (%) T oolAlpaca-4, Llama-...

  2. [12]

    Y . Meng, M. Xia, and D. Chen. SimPO: Simple preference optimization with a reference-free reward.arXiv preprint arXiv:2405.14734,

    arXiv

  3. [13]

    Mohri and T

    C. Mohri and T. Hashimoto. Language models with conformal factuality guarantees. arXiv preprint arXiv:2402.10978,

    arXiv

  4. [14]

    Papadopoulos, K

    H. Papadopoulos, K. Proedrou, V . V ovk, and A. Gammerman. Inductive confidence machines for regression. In Machine learning: ECML 2002: 13th European conference on machine learning Helsinki, Finland, August 19–23, 2002 proceedings

    2002

  5. [16]

    J. Su, J. Luo, H. Wang, and L. Cheng. Api is enough: Conformal prediction for large language models without logit-access. arXiv preprint arXiv:2403.01216,

    arXiv

  6. [17]

    Q. Tang, Z. Deng, H. Lin, X. Han, Q. Liang, B. Cao, and L. Sun. Toolalpaca: Generalized tool learning for language models with 3000 simulated cases. arXiv preprint arXiv:2306.05301,

    arXiv

  7. [19]

    Ulmer, C

    D. Ulmer, C. Zerva, and A. Martins. Non-exchangeable conformal language generation with nearest neighbors. In Findings of the Association for Computational Linguistics: EACL 2024,

    2024

  8. [21]

    F. Ye, M. Yang, J. Pang, L. Wang, D. F. Wong, E. Yilmaz, S. Shi, and Z. Tu. Benchmarking llms via uncertainty quantification. arXiv preprint arXiv:2401.12794,

    arXiv

  9. [22]

    In Appendix A.1 we provide details of LLM inference for MCQs

    13 Prune ’n Predict: Optimizing LLM Decision-making with Conformal Prediction Supplementary Material The supplementary material is organized as follows. In Appendix A.1 we provide details of LLM inference for MCQs. Appendix B provides proofs of the propositions in the paper. A...

    2023

  10. [23]

    test sample

    on x to produce a set of output logits: l1, l2, . . . ,ln ← LLM t1, t2, . . . tn (10) Here, each logit lj ∈ R|V | expresses the likelihood of the next token after t1, . . . , tj, where V is the universal set of tokens (aka the alphabet) for the given LLM and |V | is its size. ...

    2022

  11. [25]

    and deferrals to human-in-the-loop (Tailor et al., 2024; Vishwakarma et al.,

    2024

  12. [26]

    and more broadly selective prediction (El-Yaniv and Wiener, 2010; Fisch et al., 2022; Vishwakarma et al.,

    2010

  13. [2002]

    C. Qu, S. Dai, X. Wei, H. Cai, S. Wang, D. Yin, J. Xu, and J.-R. Wen. Tool learning with large language models: A survey. arXiv preprint arXiv:2405.17935,

    arXiv

  14. [2005]

    Y . Wang, X. Ma, G. Zhang, Y . Ni, A. Chandra, S. Guo, W. Ren, A. Arulraj, X. He, Z. Jiang, et al. Mmlu-pro: A more robust and challenging multi-task language understanding benchmark. In The Thirty-eight Conference on Neural Information Processing Systems Datasets and Benchmar...

    arXiv

  15. [2019]

    Touvron, L

    H. Touvron, L. Martin, K. Stone, P. Albert, A. Almahairi, Y . Babaei, N. Bashlykov, S. Batra, P. Bhargava, S. Bhosale, et al. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288,

    arXiv

  16. [2021]

    Kiyani, G

    S. Kiyani, G. Pappas, and H. Hassani. Length Optimization in Conformal Prediction. arXiv preprint arXiv:2406.18814,

    arXiv

  17. [2022]

    A. N. Angelopoulos, S. Bates, E. J. Cand`es, M. I. Jordan, and L. Lei. Learn then test: Calibrating predictive algorithms to achieve risk control. arXiv preprint arXiv:2110.01052,

    arXiv

  18. [2023]

    Kumar, C

    B. Kumar, C. Lu, G. Gupta, A. Palepu, D. Bellamy, R. Raskar, and A. Beam. Conformal prediction with large language models for multi-choice question answering. arXiv preprint arXiv:2305.18404,

    arXiv

  19. [2024]

    A. N. Angelopoulos and S. Bates. A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantification. arXiv preprint arXiv:2107.07511, (arXiv:2107.07511),

    arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.