Pith. sign in

REVIEW 3 major objections 5 minor 53 references

Interpreting Black-Box Large Language Models with Sentence-Level Energy Landscapes

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Contrastive energy modeling of prompt-response pairs lets a lightweight interpreter attribute a black-box LLM's outputs to the prompt sentences that drove them, with no further API calls after training.

desk verdict A coherent, honestly written two-stage EBM-to-interpreter attribution method for black-box LLMs; the causal faithfulness claim rests on a cosine-similarity metric that likely rewards topical overlap, so it needs a strong revision. read the letter →

arxiv 2608.02879 v1 pith:56BILXMJ submitted 2026-08-03 cs.AI

Maryam Rezaee , Pooriya Safaei , Maryam Asgarinezhad , Fatemeh Seyyedsalehi This is my paper · ORCID
classification cs.AI
keywords energy-basedmodelsblack-boxLLMinterpretabilitysentence-levelattributionpost-hocexplanationcontrastivelearningGumbel-Softmaxselectioncausalfaithfulnesssurrogatemodel
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 claims that a black-box LLM can be attributed at the sentence level by first training an energy-based surrogate over prompt-response pairs and then training a lightweight interpreter to select the prompt sentences whose removal changes the surrogate's energy most. The interpreter is meant to be standalone: after a one-time training phase on 20,000 pairs from the target LLM, it can attribute a new prompt without making any further API calls. The authors report that this attribution approaches the causal faithfulness of a sentence-level LIME baseline (Gap 0.195 versus 0.245) while running in $O(1)$ inference time, and that it breaks even with LIME on API cost after roughly 400 interpreted prompts. The pay-off, if the claims hold, is a model-agnostic post-hoc explanation tool for proprietary models that are otherwise only reachable through closed APIs.

What carries the argument

The load-bearing object is the contrastively trained energy function $E_{\mathrm{LM}}(x,y;\theta)$, a transformer-based EBM that maps sentence-embedded prompts and responses to a scalar compatibility score. It serves as a differentiable surrogate for the black-box LLM, turning the discrete search over prompt-sentence subsets into a smooth optimization: the interpreter learns to pick the mask that maximizes the energy gap between the removed and retained sentences. The training combines a fidelity InfoNCE term with a local-dependency term built from two partial-sequence samplers, and the interpreter's discrete selection is made differentiable with a Gumbel-Softmax relaxation. This machinery is what lets an energy landscape act as the supervision signal for attribution.

What would settle it

Train the same two-stage interpreter on an open-weights LLM for which internal gradients or activations are available, then compare ESCI's selected prompt sentences with gradient- or activation-based attributions on 500 counterfactual triplets; if the agreement is near chance, the energy landscape would be shown to encode something other than the target's causal computation.

Watch

Extended reading notes

Core claim

The central claim, on the paper's own terms, is that the generation behavior of an opaque LLM can be compressed into a scalar energy landscape over sentences, and that an interpreter trained to maximize the energy gap between selected and unselected prompt sentences inherits the causal structure of that landscape. The method defines each sentence as a concept, embeds prompt and response sentences with a frozen Sentence-BERT model, and passes them through transformer blocks that project into a task-adapted concept space. A contrastive dual objective—global fidelity against human or other-LM responses, plus local dependency against mismatched partial prompts and responses—shapes the energy so that low energy marks authentic target-LLM pairs. Given a target output sentence, the interpreter produces a sparse binary mask over prompt sentences by optimizing the difference between the energy of the complement and the energy of the selected subset. The experiments show that these selections align with five LLM oracles on nDCG and soft top-1 metrics and, on generative sufficiency and comprehensiveness, come close to a much more expensive LIME baseline.

Load-bearing premise

The whole method depends on the trained energy scores matching the target LLM's real cause-and-effect structure over prompt and response sentences, not just its surface style; the paper says its results confirm behavioral simulation, not mechanistic alignment.

Editorial extensions

If this is right

  • Attribution for closed-API models becomes a one-time training cost rather than a per-query cost, so high-volume monitoring of proprietary LLM outputs becomes feasible.
  • Sentence-level attributions align with rationale-evaluation standards, giving auditors units they can read and verify rather than token-level salience maps.
  • If the energy gap truly tracks causal influence, the same interpreter can be used to prune irrelevant prompt sentences before generation, reducing token and API costs without degrading the answer.
  • The faithfulness results imply that a roughly 71-million-parameter trainable interpreter can rival instruction-tuned LLM judges and a perturbation-based baseline on standard Q&A-style tasks.
  • Because the interpreter is trained across a distribution of prompts rather than a single neighborhood, attributions should generalize more reliably across inputs of the same task type.

Reading between the lines

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

  • Editorial inference: the same energy landscape is a natural substrate for auditing chain-of-thought: if an intermediate reasoning step is unfaithful, its energy relative to the prompt and final answer should be anomalously high; the paper lists this as future work but does not test it.
  • Editorial inference: the reported break-even at 400 prompts means the method is a bet on deployment volume; for one-off explanations, paying LIME's per-query cost is likely the rational choice, and the paper's own numbers do not resolve this trade-off.
  • Editorial inference: the plausibility comparison uses other LLMs as oracles, so shared stylistic biases among LLMs could inflate the alignment scores; a check against white-box explanations on an open-weights model would arbitrate this, as the paper itself notes in its limitations.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes a post-hoc, model-agnostic sentence-level attribution method for black-box LLMs. It first trains a transformer-based energy-based model (EBM) on 20,000 prompt-response pairs with a hybrid InfoNCE objective, then trains a lightweight interpreter network whose selection mask maximizes the EBM energy gap between complement and selected prompt subsets (Eq. 3). The trained interpreter is claimed to identify the prompt sentences causally responsible for a user-specified target output, with zero additional LLM API calls at inference. Experiments include an ablation of the EBM objective, alignment of interpreter scores with five LLM 'oracles', and a causal faithfulness study in which the target LLM is queried with selected and removed sentence subsets, measured by embedding cosine similarity.

Significance. If the faithfulness claim were established, the framework would offer a practical zero-query-at-inference sentence-level attribution tool for closed APIs. The paper does several things right: the faithfulness experiment in Section 4.3 is a genuine interventional test against the target LLM, not only against the surrogate; the limitations section explicitly concedes that the results support behavioral simulation rather than mechanistic alignment; and the zero-API-at-inference property is clearly stated and useful. However, the central causal claim is not yet supported by the evidence as presented, because the faithfulness metric is embedding similarity and no control rules out a purely topical selector.

major comments (3)
  1. [Section 4.3, Table 2] The central faithfulness claim rests on Generative Sufficiency/Comprehensiveness measured as cosine similarity between the embedded target sentence and the LLM's counterfactual response (Eqs. G4-G5). This metric rewards topical or stylistic overlap, not causal dependence. A selector that simply picks the prompt sentences most embedding-similar to the target sentence would likely produce a positive gap, so the reported Gap of 0.195 for ESCI versus 0.245 for LIME does not by itself demonstrate causal attribution. Please add a control baseline that selects sentences by maximum SBERT cosine similarity to the target sentence (or an equivalent topical selector), and report its Gap in Table 2. Also provide bootstrap confidence intervals or significance tests over the 200 prompts, since the ESCI/LIME difference may be within sampling noise.
  2. [Section 3.3, Eq. (3); Section 6] The interpreter is optimized against the EBM energy gap of Eq. (3), so its behavior is only as valid as the EBM's approximation of the target LLM. The EBM is trained contrastively on 20,000 observational pairs, and the paper's own Section 6 states that 'our current results confirm behavioral simulation rather than mechanistic alignment.' Appendix E.1's diagnostic evaluates the EBM using its own energy drops, not against target-LLM interventions. The abstract's claim that 'our EBM accurately simulates the target LLM' is therefore stronger than the evidence. Please either soften the abstract and introduction accordingly, or add a direct validation in which EBM energy changes on ablated prompts are compared with changes in the target LLM's actual generations (e.g., semantic similarity or likelihood under the LLM).
  3. [Section 4.2] The plausibility evaluation compares against five LLM 'oracles' rather than human judgments, and the authors acknowledge this is not ground truth. The qualitative case studies are said to reflect manual analysis by two evaluators of all 2,000 combinations, but no annotation protocol, inter-annotator agreement, or random sample details are given. Please report at least Cohen's kappa on a held-out subset, or restrict the claims to 'alignment with LLM preferences' rather than 'plausibility.'
minor comments (5)
  1. [Section 3.3 and 4.2] The acronym ESCI is used without definition before Section 4.2; define it at first introduction of the interpreter (Section 3.3).
  2. [Figure 4, sample ID 33] The sentence indices shown are [0], [1], [3]; [2] is missing, and the score columns are not aligned with the number of prompt sentences. Please fix the snippet or annotation.
  3. [Table 4] Table 4 omits the value of K (number of Gumbel-Softmax draws) used by the interpreter; Appendix C states K independent samples are drawn, but no value is reported in Table 4.
  4. [Appendix G, Eq. (G2)] The kernel width sigma=0.25*sqrt(n) is stated, but the regularization strength for the Ridge regression in LIME is not reported; this is needed to reproduce the LIME baseline.
  5. [Section 4.1, Table 1] Several metrics (e.g., Art. Delta E) are reported without variance or significance information; adding confidence intervals would make the ablation validation more robust.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: faithfulness is evaluated by external LLM interventions, not by the training objective.

full rationale

The derivation chain is self-contained against an external benchmark. The interpreter is trained to maximize the EBM energy gap (Eq. 3), but the paper's core claim of causal faithfulness is tested in Section 4.3 by actually querying the target LLM with the selected subset and its complement and measuring cosine similarity of the resulting responses to the target sentence (Table 2, Eqs. G4-G5). That evaluation is not the training objective and is not a function of the EBM's energy; it is an independent behavioral check. The EBM's ablation diagnostics in Section 4.1 and Appendix E partly use energy-drop importance derived from the EBM itself, but those are internal sanity checks, not the load-bearing evidence for the central claim; the external faithfulness benchmark is. The only self-citation (Seyyedsalehi et al., 2024) appears in Related Work as background and is not load-bearing. Section 6's concession that the results 'confirm behavioral simulation rather than mechanistic alignment' is an honest limitation statement, and the paper does not reduce any prediction to its input by construction. No circular step meets the required bar of exhibiting Eq. X = Eq. Y or a fitted parameter renamed as a prediction.

Assumptions & free parameters 8 free parameters · 5 assumptions · 1 invented entities

The central claim rests on several hand-set hyperparameters (lambda, tau, margin, K, Nmax, Gumbel temperature) and on domain assumptions about sentence-level concepts, embedding similarity, oracle LLM validity, and the faithfulness metric. No external physical entities are introduced, but the internal concept space carries much of the explanatory weight without independent validation.

free parameters (8)
  • lambda (loss weight) = 0.9 (hybrid)
    Weights fidelity vs local dependency in Eq. (1); selected by hand, not by validation.
  • InfoNCE temperature tau = 0.1
    Used in the energy scoring term h=exp(-E/tau); chosen hyperparameter.
  • margin = 0.5
    Margin in EBM training (Table 3), hand-set.
  • negative candidates K = 5
    Number of negative samples per positive in InfoNCE (Table 3).
  • Nmax (max sentence count) = 16
    Task-dependent sequence length with learnable padding; hand-set.
  • Gumbel temperature = 1.0
    Interpreter selection temperature (Table 4).
  • Grounding interval N_ground = not specified
    Period of alternating EBM fine-tuning (App. D.1); not reported.
  • Gumbel top-K = not specified
    Number of concurrently selected sentences in Appendix C; not reported in hyperparameter tables.
assumptions (5)
  • domain assumption Sentences are the appropriate operational unit of 'concept' for attribution.
    Section 1, citing Frege and Kintsch; a philosophical and linguistic premise, not established for LLM causal structure.
  • domain assumption Sentence-BERT embeddings preserve semantic similarity needed for the concept space.
    Section 3.1; frozen all-mpnet-base-v2 is assumed to be a valid semantic space for both retrieval and faithfulness evaluation.
  • domain assumption High-capacity LLMs share architectural inductive biases with the target, making them valid attribution oracles.
    Section 4.2; explicitly stated proxy assumption, conceded in Section 6.
  • domain assumption Cosine similarity between sentence embeddings is a valid measure of semantic retention for faithfulness.
    Appendix G; chosen after NLI metrics produced low scores; this assumption is load-bearing for Table 2.
  • ad hoc to paper InfoNCE contrastive learning with the chosen negative samplers shapes an energy landscape aligned with the target's input-output dynamics.
    Section 3.2 and Appendix B; the design of the local dependency samplers is specific to this paper and not a standard background result.
invented entities (1)
  • Sentence-level concept space (C_in, C_out)
    purpose: Latent space where distances reflect the LLM's internal dependency structure; used by both EBM and interpreter.
    Learned latent representation internal to the model; no external falsifiable handle beyond the paper's own downstream evaluations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Interpreting Black-Box Large Language Models with Sentence-Level Energy Landscapes." pith.science (2026). https://pith.science/paper/56BILXMJ

@misc{pith2026260802879,
  author       = {Pith},
  title        = {Pith review of: Interpreting Black-Box Large Language Models with Sentence-Level Energy Landscapes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/56BILXMJ}},
  note         = {Machine review of arXiv:2608.02879}
}
read the original abstract

The widespread adoption of proprietary Large Language Models (LLMs) accessed strictly through closed APIs has created a critical challenge for responsible deployment: a fundamental lack of interpretability. To address this, we propose a model-agnostic, post-hoc attribution interpreter operating at the sentence level. Our approach trains an Energy-Based Model (EBM) as a surrogate to capture the LLM's internal conceptual consistency between prompts and responses. This energy landscape guides the training of a lightweight interpreter network. Uniquely, our interpreter operates as a standalone tool; once trained, it quantifies the influence of prompt sentences on a user-specified target output without requiring further API queries to the LLM. By globally training a local interpreter across diverse inputs, our framework captures broader generation patterns and mitigates instance-specific biases. Experiments demonstrate that our EBM accurately simulates the target LLM, allowing the interpreter to effectively identify the prompt sentences most influential in generating specific target outputs.

Figures

Figures reproduced from arXiv: 2608.02879 by the authors.

Figure 1
Figure 1. Overview of the Proposed Framework. The prompt and response of a black-box LLM are split into sentences and embedded via a pre-trained model. An encoding module maps these embeddings to a concept space where proximity reflects relevance. Subsequently, an interaction module evaluates the consistency between input and output concepts to identify the most influential prompt sentences. These modules are trained using si… view at source ↗
Figure 2
Figure 2. Architectural Overview. Schematics for (a) the energy function ELM and (b) the interpreter network IN . 3 Methodology Our goal is to develop a post-hoc, model-agnostic method for interpreting black-box LLMs, specif￾ically by identifying which input sentences drive the response. We depart from standard token-level attribution by establishing the sentence as the fun￾damental unit of analysis. As the smallest linguisti… view at source ↗
Figure 3
Figure 3. Confusion Matrices Evaluating Interpretation Plausibility. Each cell (i, j) represents how well the Interpreter in row i matches the scores of the Oracle in column j for interpreting our target LLM. Soft Top-1 (left) measures if the Interpreter’s top choice appears in the Oracle’s top-2. nDCG (right) measures ranking correlation. Our proposed ESCI model shows remarkable alignment with GPT-J and GPT-2, suggesting it … view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Qualitative Comparison of Attribution Scores. Left: Prompt snippets. Middle: Attribution scores from ESCI and oracles. Right: Target sentence from the LLM response. ID 6: ESCI correctly attributes the simple tone to the style instruction, while oracles focus on the sub…
Figure 5
Figure 5. Figure 5: Computational Scalability. (Left) Cumula￾tive target LLM API calls required as the number of interpreted prompts scales. (Right) ESCI eliminates target API queries during inference entirely, enabling high-volume deployability. for faithfulness through exhaustive local …
Figure 6
Figure 6. Figure 6: Pre-training Pipeline of the EBM. The ar￾chitecture projects SentenceBERT embeddings into a dynamic concept space via self-attention, followed by a cross-attention mechanism to model input-output inter￾actions. An MLP aggregates these features to compute a scalar energ…
Figure 7
Figure 7. Figure 7: Overview of the Alternating Optimization Protocol. The framework employs a joint training strat￾egy to prevent distribution shift. (Left) In the standard phase, the interpreter generates a binary mask over the prompt sentences; its parameters are updated to maxi￾mize t…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 29 canonical work pages

  1. [1]

    Second Conference on Language Modeling , year=

    Energy-Based Reward Models for Robust Language Model Alignment , author=. Second Conference on Language Modeling , year=

    work page

  2. [2]

    Proceedings of the 35th International Conference on Machine Learning , pages=

    Learning to Explain: An Information-Theoretic Perspective on Model Interpretation , author=. Proceedings of the 35th International Conference on Machine Learning , pages=. 2018 , organization=

    work page 2018

  3. [3]

    The Thirteenth International Conference on Learning Representations , year=

    Concept Bottleneck Large Language Models , author=. The Thirteenth International Conference on Learning Representations , year=

    work page

  4. [4]

    The Thirteenth International Conference on Learning Representations , year=

    Scaling and evaluating sparse autoencoders , author=. The Thirteenth International Conference on Learning Representations , year=

  5. [5]

    The Thirteenth International Conference on Learning Representations , year=

    Energy-Based Diffusion Language Models for Text Generation , author=. The Thirteenth International Conference on Learning Representations , year=

  6. [6]

    2025 , eprint=

    Decomposition-Enhanced Training for Post-Hoc Attributions In Language Models , author=. 2025 , eprint=

    work page 2025

  7. [7]

    2026 , eprint=

    Latent Debate: A Surrogate Framework for Interpreting LLM Thinking , author=. 2026 , eprint=

    work page 2026

  8. [8]

    1991 , publisher=

    The Foundations of Arithmetic , author=. 1991 , publisher=

    work page 1991

Show all 53 references
  1. [9]

    1998 , publisher=

    Comprehension: A Paradigm for Cognition , author=. 1998 , publisher=

    1998

  2. [10]

    Predicting structured data , volume=

    A tutorial on energy-based learning , author=. Predicting structured data , volume=. 2006 , publisher=

    2006

  3. [11]

    2023 , eprint=

    How Close is ChatGPT to Human Experts? Comparison Corpus, Evaluation, and Detection , author=. 2023 , eprint=

    2023

  4. [12]

    Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) , year =

    Atanasova, Pepa and Camburu, Oana-Maria and Lioma, Christina and Lukasiewicz, Thomas and Simonsen, Jakob Grue and Augenstein, Isabelle , title =. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) , year =

  5. [13]

    Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing: System Demonstrations , year =

    Honnibal, Matthew and Montani, Ines , title =. Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing: System Demonstrations , year =

    2017

  6. [14]

    2024 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , year =

    Clemmer, Colton and Ding, Junhua and Feng, Yunhe , title =. 2024 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , year =

    2024

  7. [15]

    Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing , year =

    Feng, Shi and Wallace, Eric and Grissom II, Alvin and Iyyer, Mohit and Rodriguez, Pedro and Boyd-Graber, Jordan , title =. Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing , year =. doi:10.18653/v1/D18-1407 , url =

    2018 doi

  8. [16]

    , title =

    DeYoung, Jay and Jain, Sarthak and Rajani, Nazneen Fatema and Lehman, Eric and Xiong, Caiming and Socher, Richard and Wallace, Byron C. , title =. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics , year =. doi:10.18653/v1/2020.acl-main.40...

    doi

  9. [17]

    , title =

    Gururangan, Suchin and Swayamdipta, Swabha and Levy, Omer and Schwartz, Roy and Bowman, Samuel and Smith, Noah A. , title =. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies , year =....

    2018 doi

  10. [18]

    Cumulated gain-based evaluation of IR techniques , journal =

    J\". Cumulated gain-based evaluation of IR techniques , journal =. 2002 , volume =. doi:10.1145/582415.582418 , url =

    2002

  11. [19]

    International Conference on Learning Representations , year =

    Kaushik, Divyansh and Hovy, Eduard and Lipton, Zachary , title =. International Conference on Learning Representations , year =

  12. [20]

    Proceedings of the 34th International Conference on Machine Learning , year =

    Koh, Pang Wei and Liang, Percy , title =. Proceedings of the 34th International Conference on Machine Learning , year =

  13. [21]

    Findings of the Association for Computational Linguistics , year =

    Hsia, Jennifer and Pruthi, Danish and Singh, Aarti and Lipton, Zachary , title =. Findings of the Association for Computational Linguistics , year =. doi:10.18653/v1/2024.findings-eacl.88 , url =

    2024 doi

  14. [22]

    International Conference on Learning Representations , year =

    Jang, Eric and Gu, Shixiang and Poole, Ben , title =. International Conference on Learning Representations , year =

  15. [23]

    2019 , url =

    Bakhtin, Anton and Gross, Sam and Ott, Myle and Deng, Yuntian and Ranzato, Marc'Aurelio and Szlam, Arthur , title =. 2019 , url =. 1906.03351 , archivePrefix =

    2019 arXiv

  16. [24]

    Journal of Machine Learning Research , year =

    Bakhtin, Anton and Deng, Yuntian and Gross, Sam and Ott, Myle and Ranzato, Marc'Aurelio and Szlam, Arthur , title =. Journal of Machine Learning Research , year =

  17. [25]

    Barrault, Loïc and Duquenne, Paul-Ambroise and Elbayad, Maha and Kozhevnikov, Artyom and Alastruey, Belen and Andrews, Pierre and Coria, Mariano and Couairon, Guillaume and Costa-jussà, Marta R. and Dale, David and Elsahar, Hady and Heffernan, Kevin and Janeiro, João Maria and...

    2024 arXiv

  18. [26]

    Bhattacharyya, Sumanta and Rooshenas, Amirmohammad and Naskar, Subhajit and Sun, Simeng and Iyyer, Mohit and McCallum, Andrew , title =. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natur...

    doi

  19. [27]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    Chefer, Hila and Gur, Shir and Wolf, Lior , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =. doi:10.1109/CVPR46437.2021.00084 , url =

    2021

  20. [28]

    International Conference on Learning Representations , year =

    Deng, Yuntian and Bakhtin, Anton and Ott, Myle and Szlam, Arthur and Ranzato, Marc'Aurelio , title =. International Conference on Learning Representations , year =

  21. [29]

    Devlin, Jacob and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina , title =. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) , year =. doi:1...

    2019 doi

  22. [30]

    2017 , url =

    Doshi-Velez, Finale and Kim, Been , title =. 2017 , url =. 1702.08608 , archivePrefix =

    2017 arXiv

  23. [31]

    Findings of the Association for Computational Linguistics: ACL 2024 , year =

    Enouen, James and Nakhost, Hootan and Ebrahimi, Sayna and Arik, Sercan and Liu, Yan and Pfister, Tomas , title =. Findings of the Association for Computational Linguistics: ACL 2024 , year =. doi:10.18653/v1/2024.findings-acl.832 , url =

    2024 doi

  24. [32]

    Proceedings of the 41st International Conference on Machine Learning , year =

    Friedman, Dan and Lampinen, Andrew Kyle and Dixon, Lucas and Chen, Danqi and Ghandeharioun, Asma , title =. Proceedings of the 41st International Conference on Machine Learning , year =

  25. [33]

    2024 , url =

    Hackmann, Stefan and Mahmoudian, Haniyeh and Steadman, Mark and Schmidt, Michael , title =. 2024 , url =. 2403.03028 , archivePrefix =

    2024 arXiv

  26. [34]

    Fatemeh Seyyedsalehi and Mahdieh Soleymani Baghshah and Hamid R

    S. Fatemeh Seyyedsalehi and Mahdieh Soleymani Baghshah and Hamid R. Rabiee , booktitle=. 2024 , url=

    2024

  27. [35]

    Proceedings of the AAAI Conference on Artificial Intelligence , year =

    Hao, Yaru and Dong, Li and Wei, Furu and Xu, Ke , title =. Proceedings of the AAAI Conference on Artificial Intelligence , year =. doi:10.1609/aaai.v35i14.17533 , url =

    doi

  28. [36]

    , title =

    Jain, Sarthak and Wallace, Byron C. , title =. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) , year =. doi:10.18653/v1/N19-1357 , url =

    2019 doi

  29. [37]

    Proceedings of the 35th International Conference on Machine Learning , year =

    Kim, Been and Wattenberg, Martin and Gilmer, Justin and Cai, Carrie and Wexler, James and Viegas, Fernanda and Sayres, Rory , title =. Proceedings of the 35th International Conference on Machine Learning , year =

  30. [38]

    2017 , url =

    Li, Jiwei and Monroe, Will and Jurafsky, Dan , title =. 2017 , url =. 1612.08220 , archivePrefix =

    2017 arXiv

  31. [39]

    and Lee, Su-In , title =

    Lundberg, Scott M. and Lee, Su-In , title =. Advances in Neural Information Processing Systems , year =

  32. [40]

    Reimers, Nils and Gurevych, Iryna , title =. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) , year =. doi:10.18653/v1/D19-1410 , url =

    2019 doi

  33. [41]

    Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , year =

    Ribeiro, Marco Tulio and Singh, Sameer and Guestrin, Carlos , title =. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , year =. doi:10.1145/2939672.2939778 , url =

  34. [42]

    Artificial Intelligence Review , year =

    Schneider, Johannes , title =. Artificial Intelligence Review , year =. doi:10.1007/s10462-024-10916-x , url =

    doi

  35. [43]

    Proceedings of the 34th International Conference on Machine Learning , year =

    Shrikumar, Avanti and Greenside, Peyton and Kundaje, Anshul , title =. Proceedings of the 34th International Conference on Machine Learning , year =

  36. [44]

    International Conference on Learning Representations (Workshop Track) , year =

    Simonyan, Karen and Vedaldi, Andrea and Zisserman, Andrew , title =. International Conference on Learning Representations (Workshop Track) , year =

  37. [45]

    Proceedings of the 34th International Conference on Machine Learning , year =

    Sundararajan, Mukund and Taly, Ankur and Yan, Qiqi , title =. Proceedings of the 34th International Conference on Machine Learning , year =

  38. [46]

    Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics , year =

    Tu, Lifu and Pang, Richard Yuanzhe and Wiseman, Sam and Gimpel, Kevin , title =. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics , year =. doi:10.18653/v1/2020.acl-main.251 , url =

    2020 doi

  39. [47]

    , title =

    Turpin, Miles and Michael, Julian and Perez, Ethan and Bowman, Samuel R. , title =. Advances in Neural Information Processing Systems , year =

  40. [48]

    and Le, Quoc V and Zhou, Denny , title =

    Wei, Jason and Wang, Xuezhi and Schuurmans, Dale and Bosma, Maarten and Ichter, Brian and Xia, Fei and Chi, Ed H. and Le, Quoc V and Zhou, Denny , title =. Advances in Neural Information Processing Systems , year =

  41. [49]

    Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , year =

    Xie, Qizhe and Ma, Xuezhe and Dai, Zihang and Hovy, Eduard , title =. Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , year =. doi:10.18653/v1/P17-1088 , url =

    doi

  42. [50]

    Proceedings of the 32nd International Conference on Machine Learning , year =

    Xu, Kelvin and Ba, Jimmy and Kiros, Ryan and Cho, Kyunghyun and Courville, Aaron and Salakhudinov, Ruslan and Zemel, Rich and Bengio, Yoshua , title =. Proceedings of the 32nd International Conference on Machine Learning , year =

  43. [51]

    Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing , year =

    Yin, Kayo and Neubig, Graham , title =. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing , year =. doi:10.18653/v1/2022.emnlp-main.14 , url =

    2022 doi

  44. [52]

    Proceedings of the AAAI 2024 Workshop on Responsible Language Models (ReLM) , year =

    Zhao, Zhixue and Shan, Boxuan , title =. Proceedings of the AAAI 2024 Workshop on Responsible Language Models (ReLM) , year =

    2024

  45. [53]

    International Conference on Learning Representations , year =

    Zhou, Yongchao and Muresanu, Andrei Ioan and Han, Ziwen and Paster, Keiran and Pitis, Silviu and Chan, Harris and Ba, Jimmy , title =. International Conference on Learning Representations , year =

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.