Pith. sign in

REVIEW 3 major objections 4 minor 71 references

For TCFD climate-disclosure classification, the adaptation strategies that win on the source corpus—similarity-based retrieval and LoRA fine-tuning—lose most of their advantage when the disclosure source changes, while random few-shot examp

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 16:32 UTC pith:743SE2U7

load-bearing objection Strong, clean empirical comparison of LLM adaptation under source shift for TCFD classification; claims hold for the tested pair but the label-equivalence premise is untested. the 3 major comments →

arxiv 2607.17952 v1 pith:743SE2U7 submitted 2026-07-20 cs.CL cs.CE

What Transfers Under Source Shift? Definitions, Examples, and Fine-Tuning for Climate Disclosure Classification

Guosheng Li , Fenghui Ren , Bin Liu , Chuan Yu , Kaiying Ji , Lin Yue , Jun Shen , Sasa Qian This is my paper
classification cs.CL cs.CE
keywords source shiftdomain adaptationclimate disclosure classificationTCFDin-context learningfew-shot selectionLoRA fine-tuningdefinition prompting
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper asks which LLM adaptation strategies survive a change in disclosure source for TCFD climate-disclosure classification, where the same four labels appear in annual reports and in official short example disclosures. It studies eleven open- and closed-source LLMs with a fixed adaptation pool, comparing definitions, in-context examples selected randomly or by similarity, and LoRA fine-tuning. It finds that all strategies help on average cross-source, but the strongest in-source strategies do not transfer: similarity retrieval gains +8.0 points in macro-F1 on the source but only +0.5 on the target, and LoRA gains +14.1 but retains only +3.7, while random few-shot holds most of its smaller gain and definitions gain even more cross-source than in-source. The paper's proposed mechanism is that what transfers is semantic guidance about what the TCFD categories mean, and what is lost is source-specific decision patterns and target-to-source textual similarity; consequently, simpler is often safer under source shift.

Core claim

The paper's central claim is that in-source adaptation strength is not a reliable guide to cross-source reliability in this task. Using a controlled design that shares one adaptation pool between in-source and cross-source evaluation, it shows that strategies which fit the source corpus well—similarity-based example retrieval and LoRA fine-tuning—are the ones that lose the most when the source changes. Conversely, randomly selected few-shot examples, a weaker in-source baseline, retain their advantage more reliably, and definitions transfer most consistently. The paper attributes this to a distinction: semantic guidance that helps the model grasp what the four TCFD pillars mean transfers, wh

What carries the argument

The central device is a three-way controlled comparison: three adaptation strategies (definitions, examples, fine-tuning) injected at different points—prompt semantics, in-context selection, and parameters—evaluated on both an in-source held-out set and a cross-source target set drawn from the same shared label space but a different disclosure source. Two analytical instruments carry the argument: the source-binding plane, which plots each strategy's in-source gain against its cross-source gain and exposes which advantages sit above or below the perfect-transfer diagonal; and the definition staircase (D0 label names, D1 pillar descriptions, D2 the 11 Recommended Disclosures, D3 27 criteria),

Load-bearing premise

The comparison stands on one source-target pair—TCFD-CR annual-report paragraphs to TCFD-EX official examples—and on the assumption that the two corpora's shared four labels are interchangeably reliable; if that label reliability or that pair is not representative, the 'simpler is safer' ranking may not generalize.

What would settle it

Take a third disclosure source that shares the same four TCFD labels—for example, press releases or earnings-call transcripts—and run the same three strategies with the same adaptation pool. If similarity retrieval or LoRA fine-tuning matches or beats random few-shot and definitions on that target, the claimed in-source/cross-source rank reversal is specific to the CR-to-EX pair rather than a general property of source shift. A second check would be to relabel a sample of both corpora by human experts: if the four labels are not applied consistently across the two sources, the transfer compari

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • A strategy's in-source gain should not be used as its deployment value: the strategies that win most on the source corpus (retrieval, LoRA) overestimate their cross-source usefulness, so practitioners should not select by source performance alone.
  • Randomly selected few-shot examples are a stronger default than a tuned similarity retriever for cross-source deployment, because the retriever's assumption that textual similarity approximates label similarity weakens under shift.
  • Definitions should be matched in granularity to the target text rather than maximally detailed; for the TCFD setting, the official 11 Recommended Disclosures help most on the short canonical examples.
  • Fine-tuned models should be deployed with a definition prompt at inference time; swapping in D1 definitions recovers part of the cross-source loss without retraining, and the recovery is larger on stronger base models.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Because the cross-source evidence is a single pair (annual reports adapted, official examples evaluated), the 'simpler is safer' generalization is tentative until tested on other shift directions like press releases or earnings calls; the proxy A-distance shows the pair is nontrivial, but it does not guarantee representativeness.
  • The definition-granularity matching result suggests a practical deployment heuristic: estimate the target text's granularity (e.g., length and lexical alignment with TCFD wording) and choose the definition level accordingly; this could be automated and evaluated.
  • The weights-plus-prompt finding implies that archived fine-tuned classifiers are incompletely specified without their inference prompts; cross-source evaluations that freeze a training prompt may understate achievable transfer.
  • If the random-beats-retrieval pattern holds beyond this pair, it would challenge the widespread practice of tuning retrievers in-domain for few-shot classification in regulated disclosure analysis, where the cost of a wrong label is high.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper reframes TCFD climate-disclosure classification as a cross-source adaptation problem. It uses two corpora that share the four-pillar TCFD label space but differ in source: TCFD-CR (annual/sustainability reports) as the adaptation source and TCFD-EX (TCFD example disclosures) as the held-out cross-source target. The authors compare three adaptation strategies—definition prompts (D0–D3), in-context examples (random, TF-IDF, and dense retrieval), and LoRA fine-tuning—across eleven prompted LLMs (with LoRA on four open-weight bases). They report that all strategies yield positive average cross-source gains, but in-source strength does not transfer uniformly: similarity retrieval and LoRA gain most in-source yet lose most of their advantage cross-source, random few-shot retains its smaller advantage more reliably, and definitions transfer most consistently, with performance depending on the match between definition granularity and target text. The paper concludes that 'simpler is often safer' under source shift and offers practitioner-oriented defaults.

Significance. If the results are taken at face value, the paper makes a useful and non-obvious contribution: it documents a concrete setting where in-source adaptation performance is a misleading guide to cross-source deployment, and it does so with a carefully controlled design. Strengths include a shared adaptation pool for all strategies, an aligned k=8/n=2 comparison that isolates the selection rule (Figure 9), three-seed LoRA means, per-model transparency throughout the appendix tables, a clear definition staircase, and a public code release. The paper also measures gains against each strategy's own zero-shot baseline, avoiding the most common circularity in strategy comparisons. However, the external validity of the headline claim rests on a single source-target pair and on an unvalidated assumption that the two corpora's TCFD labels are interchangeable; the limited evidence directly supporting 'definitions transfer' is confounded by the way TCFD-EX appears to be constructed from the same recommended-disclosure wording as the D2 prompt.

major comments (3)
  1. [§3.1, Figures 2–3; Limitations] The central cross-source comparison assumes the four-pillar labels of TCFD-CR and TCFD-EX are interchangeable, but the paper provides no annotation-agreement or label-alignment evidence for either corpus. The cited reproducibility study (Calamai et al., 2025) reports pervasive annotation issues in climate-NLP benchmarks, so this is a live risk rather than a pedantic one. If the two corpora were annotated under different interpretations of 'Strategy' or 'Risk Management', the strategy ranking could reflect annotation artifacts rather than transfer. A concrete test would be to have annotators label a random sample from each corpus under the other corpus's label protocol and measure agreement, or at minimum to report the original annotation guidelines and a label-conditional error analysis. This assumption is load-bearing for the abstract's cross-source claim.
  2. [§5.2, Appendix B, Figure 12] The claim that definitions transfer most consistently is partly confounded by the construction of the target corpus. TCFD-EX is derived from the TCFD's example-disclosures list, and the D2 prompt inserts the 11 Recommended Disclosures. The paper's own case study (Figure 12, Case B) shows a TCFD-EX Strategy instance that is a near-verbatim paraphrase of Recommended Disclosure c. The apparent D2 advantage on TCFD-EX may therefore reflect lexical/template overlap between the prompt and the target corpus's construction, not a general property of semantic definitions under source shift. To support the general claim, the authors should evaluate on a target source not built from the same recommended-disclosure wording (e.g., press releases or earnings calls) or, at a minimum, show that the result survives after excluding near-verbatim examples.
  3. [§5.1, Limitations, Conclusion] The broad conclusion that 'when the source changes, simpler is often safer' rests entirely on a single source-target pair with one shift direction: long, discursive annual-report paragraphs to short, canonical TCFD example disclosures. The Limitations section honestly acknowledges the single pair, but the abstract and conclusion do not carry that caveat. The deployment sources named in the Introduction—press releases and earnings calls—could plausibly produce a different strategy ranking. Without additional target sources or a clearer re-scoping of the claims to the studied pair, the external validity of the headline finding remains unsupported.
minor comments (4)
  1. [§5.3, Table 1, Figure 8] The main retrieval table selects (k, n) as the per-model 'best-on-CR' configuration using the in-source test set, not a validation split. This can inflate in-source gains and make the cross-source drop look larger. The aligned k=8, n=2 comparison in Figure 9 supports the same qualitative conclusion, so this is not a fatal flaw, but the authors should either use validation-based selection or acknowledge the optimistic bias in Table 1.
  2. [§5.1, Table 7] The text emphasizes that similarity retrieval 'barely improves' cross-source on average, but Table 7 shows that the best TF-IDF configuration reaches the highest TCFD-EX macro-F1 of any strategy (0.713, Qwen3.5-27B), above the best definition prompt (0.707). The average statements should be accompanied by per-model ranges so readers do not conclude retrieval is uniformly poor cross-source.
  3. [Figures 5–6 and Tables 1/4/5] Most reported means lack confidence intervals or model-level variance bars; only Figure 9 provides a bootstrap CI. Since the headline claims are about average behavior across models, reporting CIs for the mean gains (or at least per-model dot plots) would make the strength of the claims easier to assess.
  4. [Abstract and §5.1] The statement 'all strategies bring positive cross-source gains on average' is true for the mean, but Figure 5 and Table 1 show several per-model points at or below zero for random and similarity few-shot. The authors should make the per-model dispersion more prominent in the prose to avoid overstating the robustness of the average.

Circularity Check

0 steps flagged

No significant circularity: all gains are measured against each strategy's own no-adaptation baseline using external corpora and models; the D2/TCFD-EX wording overlap is a disclosed benchmark-construction caveat, not a circular derivation.

full rationale

The paper is an empirical measurement study, not a derivation chain. TCFD-CR (Bingler et al., 2022) and TCFD-EX (Qiu, 2023) are external corpora; the eleven LLMs and the ClimateBERT-TCFD reference are external; and each strategy's gain is computed against its own no-adaptation baseline (definitions D2 vs D0; LoRA adapter vs base at D1; random/similarity vs D1 zero-shot). No reported cross-source number is produced by fitting to TCFD-EX, and the hyperparameters tuned on TCFD-CR (retrieval k,n; LoRA validation checkpoint) are applied to TCFD-EX without target-side selection, so the target result is not a refit of its own input. The only construction-adjacent issue is that TCFD-EX is derived from the TCFD's example-disclosures list and D2 contains the 11 Recommended Disclosures; the paper itself states that TCFD-EX instances are 'closer to the wording of the TCFD Recommended Disclosures' (Section 5.2), so the D2 advantage on that target is partly explained by overlapping source text. This is a benchmark-construction/external-validity caveat, not a circular step: the paper does not define D2's success in terms of TCFD-EX labels, does not claim a derivation from definitions to labels, and the finding is presented as an observed mechanism. The Limitations section explicitly concedes the single-pair design ('Our cross-source evaluation is built on a single dataset pair, adapting on TCFD-CR and evaluating on TCFD-EX'), which is a generalization risk rather than a circularity. No load-bearing self-citation appears in the argument; the domain-adaptation framing cites external, falsifiable work. Therefore there is no circularity to score above zero.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

The central claims rest on experimental measurements, so the ledger mostly records design choices rather than fitted constants. Three hand-chosen configuration families (retrieval (k,n), LoRA hyperparameters, definition staircase), four domain assumptions about corpus equivalence and representativeness, and no invented entities. The paper is transparent about most of these; the least-supported choice is treating TCFD-EX as representative of 'new sources' more generally, and the least-tested is label equivalence across corpora.

free parameters (3)
  • Retrieval (k, n) per-model configuration = k∈{3,5,8}, n∈{2,3}; best-on-CR per model
    Headline retrieval gains (+8.0 pp in-source, +0.5 pp cross-source, Table 1) use the (k,n) that maximizes macro-F1 on the TCFD-CR test split per model, while the random baseline is fixed at k=8,n=2; the aligned k=8,n=2 comparison (Figure 9) confirms the direction but the magnitude of the in-source gap partly reflects tuning asymmetry.
  • LoRA hyperparameters = r=16, α=32, lr=2e-4, dropout 0.05, QLoRA 4-bit, ≤5 epochs, best checkpoint on validation, 3 seeds
    Standard configuration from §4.1.3; the LoRA conclusions (large in-source gain, consistent cross-source drop, D1-prompt recovery) hold across all four bases, limiting sensitivity to these choices.
  • Definition staircase levels (D1–D3) = 4 pillar intents, 11 Recommended Disclosures, 27 criteria
    Built from official TCFD documents and Doi et al. (2024), so not invented; but the claim that 'granularity matches target text' depends on the specific staircase quantization, and is inferred from where peaks fall in it.
axioms (4)
  • domain assumption The four TCFD labels are annotated equivalently across TCFD-CR and TCFD-EX, so cross-source macro-F1 is a fair measure
    The entire evaluation treats Bingler et al. (2022) and Qiu (2023) labels as interchangeable (§3); label reliability differences between the corpora are not measured, and the paper's own related work cites Calamai et al. (2025) finding pervasive annotation issues in climate-NLP benchmarks.
  • domain assumption TCFD-EX is a representative stand-in for 'new disclosure source'
    TCFD-EX is the TCFD's own curated example disclosures — short, canonical, close to the Recommended Disclosure wording (the paper's own §5.2 and Appendix F say so); real press releases or earnings calls may resemble neither corpus. The 'simpler is often safer' generalization inherits this premise, as the Limitations section concedes the single-pair design.
  • standard math Proxy A-distance (logistic regression on sentence embeddings) measures a nontrivial shift
    Used in §3.1 (81.2% source-classifier accuracy, A-distance ≈1.25) to argue the two corpora are not near-duplicates; this is the standard Ben-David et al. (2007) framework, but separability of the corpora does not by itself establish that the shift behaves like the deployment shifts practitioners face.
  • domain assumption A single fixed prompt frame per definition level is sufficient to measure definition effects
    Only one prompt template is used (Figure 11); known LLM sensitivity to prompt wording is not tested, so part of the definition-level differences could be wording-specific rather than semantic-granularity effects.

pith-pipeline@v1.3.0-alltime-deepseek · 15727 in / 21929 out tokens · 165515 ms · 2026-08-01T16:32:02.356072+00:00 · methodology

0 comments
read the original abstract

Climate disclosure classification is a fundamental task for analysing corporate climate disclosures, yet such disclosures appear in many different sources -- annual reports, press releases, and earnings calls -- that differ in length, purpose, and writing style. Existing evaluations are mostly conducted within a single source, leaving open whether common LLM adaptation strategies remain effective under source shift. We reframe climate disclosure classification as a cross-source adaptation problem and study three widely used adaptation strategies -- definitions, examples, and fine-tuning -- across eleven open- and closed-source LLMs, using two corpora that share the same label space but come from different sources. We find that all strategies bring positive cross-source gains on average, but the strongest in-source strategies are not the strongest cross-source ones: similarity-based retrieval and LoRA fine-tuning gain most in-source but lose most of that advantage under source shift; randomly selected few-shot examples, a weaker in-source baseline, retain their advantage more reliably; definitions transfer most consistently, though only when their granularity matches the target text. Across these strategies, when the source changes, simpler is often safer.

Figures

Figures reproduced from arXiv: 2607.17952 by Bin Liu, Chuan Yu, Fenghui Ren, Guosheng Li, Jun Shen, Kaiying Ji, Lin Yue, Sasa Qian.

Figure 1
Figure 1. Figure 1: Deployment gap motivating this study. Re [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: quantifies the size, style, and class con￾trast between the two corpora [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: A logistic-regression source classifier distin [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Overview of the experimental design: three [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Each adaptation strategy as a point in the in [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Definition specificity under zero-shot prompt [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: TF-IDF retrieval gain over zero-shot, per [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 10
Figure 10. Figure 10: Per-base LoRA trajectory across in-source (TCFD-CR) and cross-source (TCFD-EX) evaluation, with [PITH_FULL_IMAGE:figures/full_fig_p008_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: A single prompt instance under the definition staircase. The frame, target instance, and output constraint [PITH_FULL_IMAGE:figures/full_fig_p013_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Two Qwen3-32B zero-shot predictions illustrating the definition peak-shift mechanism. Case A: a noisy [PITH_FULL_IMAGE:figures/full_fig_p014_12.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

71 extracted references · 1 canonical work pages

  1. [1]

    Alix Auzepy, Elena T \"o njes, David Lenz, and Christoph Funk. 2023. https://doi.org/10.1371/journal.pone.0288052 Evaluating TCFD reporting---a new application of zero-shot analysis to climate-related financial disclosures . PLOS ONE, 18(11):e0288052

    work page doi:10.1371/journal.pone.0288052 2023

  2. [2]

    Shai Ben-David, John Blitzer, Koby Crammer, and Fernando Pereira. 2007. Analysis of representations for domain adaptation. In Advances in Neural Information Processing Systems (NeurIPS), pages 137--144

    2007

  3. [3]

    Julia Anna Bingler, Mathias Kraus, Markus Leippold, and Nicolas Webersinke. 2022. https://doi.org/10.1016/j.frl.2022.102776 Cheap talk and cherry-picking: What ClimateBERT has to say on corporate climate risk disclosures . Finance Research Letters, 47:102776

    arXiv 2022

  4. [4]

    Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, and 1 others

    Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, and 1 others. 2020. Language models are few-shot learners. In Advances in Neural Information Processing Systems, volume 33, pages 1877--1901

    2020

  5. [5]

    Suchanek

    Tom Calamai, Oana Balalau, and Fabian M. Suchanek. 2025. https://aclanthology.org/2025.findings-acl.925/ Benchmarking the benchmarks: Reproducing climate-related NLP tasks . In Findings of the Association for Computational Linguistics: ACL 2025. Association for Computational Linguistics

    2025

  6. [6]

    Nitay Calderon, Naveh Porat, Eyal Ben-David, Alexander Chapanin, Zorik Gekhman, Nadav Oved, Vitaly Shalumov, and Roi Reichart. 2024. https://aclanthology.org/2024.findings-emnlp.7/ Measuring the robustness of NLP models to domain shifts . In Findings of the Association for Computational Linguistics: EMNLP 2024. Association for Computational Linguistics

    2024

  7. [7]

    Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, and Luke Zettlemoyer. 2023. QLoRA : Efficient finetuning of quantized LLMs . In Advances in Neural Information Processing Systems, volume 36

    2023

  8. [8]

    Korenari Doi and 1 others. 2024. https://www.jpx.co.jp/english/corporate/research-study/working-paper/ Automated determination of TCFD recommended disclosures through zero-shot text classification using large language models . JPX Working Paper Vol. 43, Japan Exchange Group. Joint research with Sherpa & Company

    2024

  9. [9]

    Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Jingyuan Ma, Rui Li, Heming Xia, Jingjing Xu, Zhiyong Wu, Tianyu Liu, Baobao Chang, Xu Sun, Lei Li, and Zhifang Sui. 2024. A survey on in-context learning. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics

    2024

  10. [10]

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, and 1 others. 2024. https://arxiv.org/abs/2407.21783 The L lama 3 herd of models . arXiv preprint arXiv:2407.21783

    Pith/arXiv arXiv 2024

  11. [11]

    Ishaan Gulrajani and David Lopez-Paz. 2021. In search of lost domain generalization. In International Conference on Learning Representations

    2021

  12. [12]

    Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. LoRA : Low-rank adaptation of large language models. In International Conference on Learning Representations

    2022

  13. [13]

    Goucher, Adam Perelman, and 1 others

    Aaron Hurst, Adam Lerer, Adam P. Goucher, Adam Perelman, and 1 others. 2024. https://arxiv.org/abs/2410.21276 GPT-4o system card . arXiv preprint arXiv:2410.21276

    Pith/arXiv arXiv 2024

  14. [14]

    IFRS Foundation . 2023. IFRS S2 climate-related disclosures. IFRS Sustainability Disclosure Standard

    2023

  15. [15]

    Earnshaw, Imran S

    Pang Wei Koh, Shiori Sagawa, Henrik Marklund, Sang Michael Xie, Marvin Zhang, Akshay Balsubramani, Weihua Hu, Michihiro Yasunaga, Richard Lanas Phillips, Irena Gao, Tony Lee, Etienne David, Ian Stavness, Wei Guo, Berton A. Earnshaw, Imran S. Haque, Sara Beery, Jure Leskovec, Anshul Kundaje, and 4 others. 2021. WILDS : A benchmark of in-the-wild distributi...

    2021

  16. [17]

    Jiachang Liu, Dinghan Shen, Yizhe Zhang, Bill Dolan, Lawrence Carin, and Weizhu Chen. 2022. What makes good in-context examples for GPT-3 ? In Proceedings of Deep Learning Inside Out: The 3rd Workshop on Knowledge Extraction and Integration for Deep Learning Architectures. Association for Computational Linguistics

    2022

  17. [18]

    Sourab Mangrulkar, Sylvain Gugger, Lysandre Debut, Younes Belkada, Sayak Paul, and Benjamin Bossan. 2022. https://github.com/huggingface/peft PEFT : State-of-the-art parameter-efficient fine-tuning methods . GitHub repository, HuggingFace

    2022

  18. [19]

    Meta AI . 2025. https://ai.meta.com/blog/llama-4-multimodal-intelligence/ The L lama 4 herd: The beginning of a new era of natively multimodal AI innovation . Blog post

    2025

  19. [20]

    Mistral AI . 2024. https://mistral.ai/news/ministraux/ Un ministral, des ministraux: The world's best edge models . Blog post

    2024

  20. [21]

    Marius Mosbach, Tiago Pimentel, Shauli Ravfogel, Dietrich Klakow, and Yanai Elazar. 2023. https://aclanthology.org/2023.findings-acl.779/ Few-shot fine-tuning vs.\ in-context learning: A fair comparison and evaluation . In Findings of the Association for Computational Linguistics: ACL 2023, pages 12284--12314. Association for Computational Linguistics

    2023

  21. [22]

    Keane Ong, Rui Mao, Deeksha Varshney, Erik Cambria, and Gianmarco Mengaldo. 2025. Towards robust ESG analysis against greenwashing risks: Aspect-action analysis with cross-category generalization. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics. Association for Computational Linguistics

    2025

  22. [24]

    Keqin Peng, Liang Ding, Yancheng Yuan, Xuebo Liu, Min Zhang, Yuanxin Ouyang, and Dacheng Tao. 2024. https://aclanthology.org/2024.acl-long.492/ Revisiting demonstration selection strategies in in-context learning . In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL). Association for Computational Linguistics

    2024

  23. [25]

    Rui Qiu. 2023. TCFD\_disclosure : A dataset of TCFD -labelled example disclosures. Hugging Face dataset. https://huggingface.co/datasets/rexarski/TCFD_disclosure

    2023

  24. [26]

    Qwen Team . 2026. https://qwen.ai/blog?id=qwen3.5 Qwen3.5 : Towards native multimodal agents . Blog post

    2026

  25. [27]

    Nils Reimers and Iryna Gurevych. 2019. https://aclanthology.org/D19-1410/ Sentence- BERT : Sentence embeddings using S iamese BERT -networks . In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). Association for Computational Linguistics

    2019

  26. [28]

    Dmitri Roussinov, Serge Sharoff, and Nadezhda Puchnina. 2025. Controlling out-of-domain gaps in LLMs for genre classification and generated text detection. In Proceedings of the 31st International Conference on Computational Linguistics (COLING). Association for Computational Linguistics

    2025

  27. [29]

    Ohad Rubin, Jonathan Herzig, and Jonathan Berant. 2022. Learning to retrieve prompts for in-context learning. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Association for Computational Linguistics

    2022

  28. [30]

    Bach, Lintang Sutawika, Zaid Alyafeai, Antoine Chaffin, Arnaud Stiegler, Teven Le Scao, Arun Raja, and 1 others

    Victor Sanh, Albert Webson, Colin Raffel, Stephen H. Bach, Lintang Sutawika, Zaid Alyafeai, Antoine Chaffin, Arnaud Stiegler, Teven Le Scao, Arun Raja, and 1 others. 2022. https://arxiv.org/abs/2110.08207 Multitask prompted training enables zero-shot task generalization . In International Conference on Learning Representations (ICLR)

    Pith/arXiv arXiv 2022

  29. [31]

    Tobias Schimanski, Julia Bingler, Camilla Hyslop, Mathias Kraus, and Markus Leippold. 2023. ClimateBERT-NetZero : Detecting and assessing net zero and reduction targets. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics

    2023

  30. [32]

    Dominik Stammbach, Nicolas Webersinke, Julia Anna Bingler, Mathias Kraus, and Markus Leippold. 2023. Environmental claim detection. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers). Association for Computational Linguistics

    2023

  31. [33]

    Task Force on Climate-related Financial Disclosures . 2017. Recommendations of the task force on climate-related financial disclosures. Report

    2017

  32. [35]

    Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M

    Jason Wei, Maarten Bosma, Vincent Y. Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M. Dai, and Quoc V. Le. 2022. https://arxiv.org/abs/2109.01652 Finetuned language models are zero-shot learners . In International Conference on Learning Representations (ICLR)

    Pith/arXiv arXiv 2022

  33. [36]

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, and 1 others. 2025. https://arxiv.org/abs/2505.09388 Qwen3 technical report . arXiv preprint arXiv:2505.09388

    Pith/arXiv arXiv 2025

  34. [37]

    An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, and 1 others. 2024. https://arxiv.org/abs/2412.15115 Qwen2.5 technical report . arXiv preprint arXiv:2412.15115

    Pith/arXiv arXiv 2024

  35. [38]

    Advances in Neural Information Processing Systems (NeurIPS) , pages =

    Ben-David, Shai and Blitzer, John and Crammer, Koby and Pereira, Fernando , title =. Advances in Neural Information Processing Systems (NeurIPS) , pages =

  36. [39]

    2110.12010 , archivePrefix =

    Webersinke, Nicolas and Kraus, Mathias and Bingler, Julia Anna and Leippold, Markus , year =. 2110.12010 , archivePrefix =

    Pith/arXiv arXiv

  37. [40]

    Cheap Talk and Cherry-Picking: What

    Bingler, Julia Anna and Kraus, Mathias and Leippold, Markus and Webersinke, Nicolas , journal =. Cheap Talk and Cherry-Picking: What. 2022 , doi =

    2022

  38. [41]

    2023 , publisher =

    Schimanski, Tobias and Bingler, Julia and Hyslop, Camilla and Kraus, Mathias and Leippold, Markus , booktitle =. 2023 , publisher =

    2023

  39. [42]

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

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

  40. [43]

    Towards Robust

    Ong, Keane and Mao, Rui and Varshney, Deeksha and Cambria, Erik and Mengaldo, Gianmarco , booktitle =. Towards Robust. 2025 , publisher =

    2025

  41. [44]

    2017 , institution =

    Recommendations of the Task Force on Climate-related Financial Disclosures , author =. 2017 , institution =

    2017

  42. [45]

    2023 , institution =

    2023

  43. [46]

    Evaluating

    Auzepy, Alix and T. Evaluating. PLOS ONE , volume =. 2023 , doi =

    2023

  44. [47]

    The Journal of Finance , volume =

    Firm-Level Climate Change Exposure , author =. The Journal of Finance , volume =. 2023 , doi =

    2023

  45. [48]

    Controlling Out-of-Domain Gaps in

    Roussinov, Dmitri and Sharoff, Serge and Puchnina, Nadezhda , booktitle =. Controlling Out-of-Domain Gaps in. 2025 , publisher =

    2025

  46. [49]

    Measuring the Robustness of

    Calderon, Nitay and Porat, Naveh and Ben-David, Eyal and Chapanin, Alexander and Gekhman, Zorik and Oved, Nadav and Shalumov, Vitaly and Reichart, Roi , booktitle =. Measuring the Robustness of. 2024 , publisher =

    2024

  47. [50]

    Few-shot Fine-tuning vs.\ In-context Learning:

    Mosbach, Marius and Pimentel, Tiago and Ravfogel, Shauli and Klakow, Dietrich and Elazar, Yanai , booktitle =. Few-shot Fine-tuning vs.\ In-context Learning:. 2023 , publisher =

    2023

  48. [51]

    , booktitle =

    Calamai, Tom and Balalau, Oana and Suchanek, Fabian M. , booktitle =. Benchmarking the Benchmarks: Reproducing Climate-Related. 2025 , publisher =

    2025

  49. [52]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL) , year =

    Revisiting Demonstration Selection Strategies in In-Context Learning , author =. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL) , year =

  50. [53]

    Automated Determination of

    Doi, Korenari and others , institution =. Automated Determination of. 2024 , month = mar, note =

    2024

  51. [54]

    and Zhang, Hao and Stoica, Ion , booktitle =

    Kwon, Woosuk and Li, Zhuohan and Zhuang, Siyuan and Sheng, Ying and Zheng, Lianmin and Yu, Cody Hao and Gonzalez, Joseph E. and Zhang, Hao and Stoica, Ion , booktitle =. Efficient Memory Management for Large Language Model Serving with. 2023 , publisher =. doi:10.1145/3600006.3613165 , url =

    arXiv 2023

  52. [55]

    Mangrulkar, Sourab and Gugger, Sylvain and Debut, Lysandre and Belkada, Younes and Paul, Sayak and Bossan, Benjamin , year =

  53. [56]

    Advances in Neural Information Processing Systems , volume =

    Language Models are Few-Shot Learners , author =. Advances in Neural Information Processing Systems , volume =

  54. [57]

    Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , year =

    A Survey on In-context Learning , author =. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , year =

    2024

  55. [58]

    What Makes Good In-Context Examples for

    Liu, Jiachang and Shen, Dinghan and Zhang, Yizhe and Dolan, Bill and Carin, Lawrence and Chen, Weizhu , booktitle =. What Makes Good In-Context Examples for. 2022 , publisher =

    2022

  56. [59]

    Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies , year =

    Learning To Retrieve Prompts for In-Context Learning , author =. Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies , year =

    2022

  57. [60]

    and Shen, Yelong and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Wang, Lu and Chen, Weizhu , booktitle =

    Hu, Edward J. and Shen, Yelong and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Wang, Lu and Chen, Weizhu , booktitle =

  58. [61]

    Dettmers, Tim and Pagnoni, Artidoro and Holtzman, Ari and Zettlemoyer, Luke , booktitle =

  59. [62]

    International Conference on Learning Representations , year =

    In Search of Lost Domain Generalization , author =. International Conference on Learning Representations , year =

  60. [63]

    and Haque, Imran S

    Koh, Pang Wei and Sagawa, Shiori and Marklund, Henrik and Xie, Sang Michael and Zhang, Marvin and Balsubramani, Akshay and Hu, Weihua and Yasunaga, Michihiro and Phillips, Richard Lanas and Gao, Irena and Lee, Tony and David, Etienne and Stavness, Ian and Guo, Wei and Earnshaw, Berton A. and Haque, Imran S. and Beery, Sara and Leskovec, Jure and Kundaje, ...

    2021

  61. [64]

    Sentence-

    Reimers, Nils and Gurevych, Iryna , booktitle =. Sentence-. 2019 , publisher =

    2019

  62. [65]

    International Conference on Learning Representations (ICLR) , year =

    Finetuned Language Models Are Zero-Shot Learners , author =. International Conference on Learning Representations (ICLR) , year =

  63. [66]

    International Conference on Learning Representations (ICLR) , year =

    Multitask Prompted Training Enables Zero-Shot Task Generalization , author =. International Conference on Learning Representations (ICLR) , year =

  64. [67]

    Grattafiori, Aaron and Dubey, Abhimanyu and Jauhri, Abhinav and Pandey, Abhinav and others , journal =. The. 2024 , url =

    2024

  65. [68]

    2025 , howpublished =

    The. 2025 , howpublished =

    2025

  66. [69]

    2024 , url =

    Yang, An and Yang, Baosong and Zhang, Beichen and Hui, Binyuan and others , journal =. 2024 , url =

    2024

  67. [70]

    2025 , url =

    Yang, An and Li, Anfeng and Yang, Baosong and Zhang, Beichen and others , journal =. 2025 , url =

    2025

  68. [71]

    2026 , howpublished =

    2026

  69. [72]

    2024 , howpublished =

    Un Ministral, des Ministraux: The World's Best Edge Models , author =. 2024 , howpublished =

    2024

  70. [73]

    and Perelman, Adam and others , journal =

    Hurst, Aaron and Lerer, Adam and Goucher, Adam P. and Perelman, Adam and others , journal =. 2024 , url =

    2024

  71. [74]

    arXiv preprint arXiv:2601.03267 , year =

    Pith/arXiv arXiv