Pith. sign in

REVIEW 3 major objections 5 minor 22 references

Causal Evidence Extraction and Triangulation in Crisis Reports using Large Language Models: A ReliefWeb-based Study

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A two-stage LLM pipeline extracts auditable causal evidence from long humanitarian reports.

desk verdict The extraction pipeline is a genuine contribution, but the headline LoE convergence score is underdetermined because a third of the extracted records are excluded from polarity normalization with no sensitivity analysis. read the letter →

arxiv 2608.04576 v1 pith:XK4YWFZR submitted 2026-08-05 cs.CL

Yuanjun Zhang , Mourad Oussalah This is my paper · ORCID
classification cs.CL
keywords causalevidenceextractionhumanitarianreportsReliefWeblargelanguagemodelsquery-conditionedsnippetgroundingtriangulationcashassistance
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that long, noisy, multi-topic humanitarian reports can be turned into structured, auditable causal evidence at scale. The proposed two-stage LLM pipeline first extracts candidate intervention–outcome pairs restricted to a query-defined intervention class, then classifies each pair's direction (increase, no change, decrease) and strength (weak, moderate, strong) using the supporting text snippet that grounds the claim. On 100 expert-annotated ReliefWeb reports covering 220 relations, the paper reports weighted F1 of 90.73% for the best closed-source model (Qwen-Plus with two-step query-conditioned, snippet-grounded extraction) and 94.15% for a LoRA fine-tuned Llama-3.1-8B student distilled from it. To synthesize across contexts, the paper introduces a cell-based triangulation score that aggregates strength-weighted evidence within disaster×source cells, applies Laplace smoothing, and averages cells equally to produce a Level-of-Evidence score; applied to cash assistance and food-related outcomes, it yields LoE = 0.865. If correct, this gives humanitarian decision-makers a scalable way to consolidate what works where from thousands of field reports, with every extracted relation traceable to a quoted snippet.

What carries the argument

The load-bearing mechanism is the two-stage extraction pipeline with query conditioning and snippet grounding. Stage 1 receives a report $x$ and a query $q$ (an intervention class, e.g., cash assistance) and extracts candidate intervention–outcome pairs $a,o$ plus a supporting snippet $z$; Stage 2, given the candidates and snippets, predicts direction $\delta\in\{\mathrm{INC},\mathrm{DEC},\mathrm{NO}\}$ and strength $\sigma\in\{\mathrm{WEAK},\mathrm{MOD},\mathrm{STRONG}\}$, producing records $(a,o,\delta,\sigma,z)$ with $z$ providing auditability. For triangulation, records are placed in cells $c=(d,s)$ by disaster type and source type; strength is mapped to ordinal weights $w(\mathrm{WEAK})=0$, $w(\mathrm{MOD})=1$, $w(\mathrm{STRONG})=2$, and each cell yields smoothed probabilities via Laplace smoothing ($\alpha=0.1$). Cells are averaged equally, and the Level-of-Evidence score is $\mathrm{LoE}=(\max(P_P,P_Z,P_N)-1/3)/(1-1/3)\in[0,1]$, measuring how much the cell-averaged evidence converges on one directional signal.

What would settle it

Have two independent experts re-annotate a random subset of the 100 evaluation reports and compute inter-annotator agreement on direction and strength labels; if agreement is low (e.g., $\kappa<0.6$), the reported 90.73% and 94.15% weighted F1 scores are not trustworthy. Alternatively, recompute the LoE for cash assistance with $\alpha=0$, with alternative polarity assignments for ambiguous outcomes such as food insecurity, and with cells weighted by report count rather than equally; if LoE drops sharply, the 0.865 convergence score is an artifact of the smoothing, polarity labels, or equal-cell-weighting design.

Watch

Extended reading notes

Core claim

The central discovery is that the over-extraction problem that plagues naive LLM relation extraction on humanitarian reports can be controlled by query-conditioning the first stage, and that grounding the second-stage relation classification in short supporting snippets improves faithfulness and auditability. The paper claims these design choices, not model size, drive accuracy: across Qwen-Plus, GPT-4o-mini, and DeepSeek-V3, query conditioning consistently yields the largest gains, and the best closed-source configuration reaches 90.73% weighted F1. It further claims that a small open-weight student (Llama-3.1-8B-Instruct) fine-tuned with LoRA on teacher outputs reaches 94.15% weighted F1, exceeding the closed-source teacher, while the distilled model remains two-stage and snippet-grounded. For synthesis, the paper's key claim is that context-preserving triangulation—aggregating within disaster×source cells, smoothing, then equally weighting cells—reveals cross-context convergence rather than pooling raw counts; cash assistance shows strong positive convergence for food-related outcomes (LoE = 0.865, $P_P=0.91$) over 19 non-empty cells, with stable long-horizon trajectories.

Load-bearing premise

Everything hinges on the 100 test reports that experts labeled by hand; the paper gives no measure of whether different experts would label them the same way, so both the F1 scores and the triangulation results could be built on uncertain ground.

Editorial extensions

If this is right

  • Query conditioning reduces over-extraction: on Qwen-Plus, weighted F1 rises from 61.44% in the two-step baseline to 89.78% with conditioning, and to 90.73% with snippet grounding added.
  • Every extracted relation carries a supporting snippet, so a human reviewer can check the evidence behind each direction and strength label without rereading the full report.
  • A small open-weight model fine-tuned on teacher outputs reaches 94.15% weighted F1, which would allow near-free deployment on much larger report corpora than the 8,029 used here.
  • The triangulation score is designed so that high-volume contexts do not dominate: equal cell weighting across 19 non-empty disaster×source cells yields $P_P=0.91$ and LoE=0.865 for cash assistance on food outcomes.
  • Cumulative temporal trajectories can monitor whether evidence convergence changes over time, e.g., peaking around 2017 and mildly softening in 2018–2024.

Reading between the lines

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

  • The same two-stage pipeline could be pointed at other intervention classes (WASH, shelter, vaccination) and other outcome families; the paper only demonstrates cash assistance, so whether the 90%+ F1 transfers is untested.
  • A natural stress test is to vary the design choices the paper fixes—Laplace $\alpha$, strength weights, and the set of polarity labels—and see how much LoE moves; the paper does not report a sensitivity analysis, and LoE=0.865 could be partly an artifact of those choices.
  • An independent re-annotation of a subset of the 100 reports with inter-annotator agreement would tell whether the reported F1 gap between closed-source and distilled student is real; the paper reports no such agreement.
  • If equal cell weighting is meant to downweight high-volume contexts, it also downweights high-quality evidence from major sources; weighting cells by source reliability or report recency would be a logical alternative the paper leaves unexplored.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper develops and evaluates a two-stage LLM extraction pipeline for ReliefWeb humanitarian reports, with query-conditioned extraction and snippet-grounded relation classification, and a cell-based triangulation method that computes a Level-of-Evidence (LoE) score. On an expert-annotated sample of 100 reports (220 relations), the best closed-source configuration (Qwen-Plus, two-step QC+SG) achieves a weighted F1 of 90.73%, and LoRA-finetuned Llama-3.1-8B achieves 94.15%. Applied to cash-assistance reports, the triangulation yields LoE=0.865 for food-related outcomes with a positive global probability P_P=0.91.

Significance. The extraction contribution is well positioned, and the evaluation design has real strengths: the benchmark is independently expert-annotated, the 100 evaluation reports are excluded from the distillation data, and the ablations (base vs QC vs QC+SG) are clean. The cost-efficiency comparison and the LoRA distillation are practically useful. If the extraction results survive a robustness analysis of the annotation reliability and the triangulation score is recomputed with a clearly specified treatment of rare outcomes and a sensitivity analysis, the work would be a valuable demonstration of auditable evidence extraction at scale. The triangulation method is transparently specified, but the headline convergence score is not yet robust because the treatment of rare outcomes is unspecified and the aggregation contains choices that can materially change the result; the limitations section acknowledges this but does not quantify it.

major comments (3)
  1. [Section 3.2, Appendix D, Eqs. (2)–(5)] The triangulation case study labels polarity only for outcome strings with frequency ≥3, which cover 13,554 of 19,568 records (69.27%). The manuscript never states how the remaining 6,014 records (5,080 rare outcome vocabularies, 30.73%) are treated in Eqs. (2)–(5). If they are dropped, LoE=0.865 is computed on a subset of the extracted evidence; if they are included without polarity normalization, positively and negatively framed rare outcomes are not compared on the same scale. Either way, the headline convergence is not a summary of the extracted evidence base, and the stress-test concern is valid. Please specify the inclusion rule for rare outcomes, report cell-level and global results on both the full and frequent-outcome subsets, and add a sensitivity analysis over the frequency threshold and the treatment of unlabeled records.
  2. [Eqs. (2)–(4)] The strength weighting w(WEAK)=0 means weak evidence never contributes to the cell counts. For example, a cell with 100 WEAK positive records and 1 MOD negative record has C_P=0 and C_N=1, and after Laplace smoothing (α=0.1) the cell probability is P_N=0.846. Since Eq. (4) gives every non-empty cell equal weight, sparse cells of this kind can move the global LoE substantially. The Limitations concede sensitivity to weighting and smoothing, but no sensitivity analysis or per-cell counts are reported. Please report the cell count matrix and vary α and the strength weights, or justify w(WEAK)=0 with an explicit audit-trail rationale that shows the robustness of the LoE=0.865 claim.
  3. [Section 3, Table 2, Appendix C] The central extraction claim rests on 100 expert-annotated reports containing 220 relations, yet no inter-annotator agreement is reported and no confidence intervals are given for the weighted F1 scores. With only 220 relations, differences such as 90.73 vs 89.17 (Qwen-Plus vs DeepSeek-V3 in the two-step QC+SG condition) may not be statistically distinguishable. The authors should report an agreement measure (e.g., Cohen's kappa or Krippendorff's alpha) and bootstrap or exact binomial confidence intervals for the headline F1 numbers, particularly because the Limitations note that performance estimates may be sensitive to sampling and annotation guidelines.
minor comments (5)
  1. [Section 4.3] The statement '87 positive vs. 65 negative outcome strings with frequency ≥3' should clarify whether these are unique outcome vocabularies or occurrence counts, since the two readings have different implications for the coverage of the polarity lexicon.
  2. [Table 2] The weights '0.30/0.30/0.20/0.20' should be mapped explicitly to the four scored components (intervention, outcome, direction, strength) in the table caption, and the partial-credit scheme for strength should be stated there as well.
  3. [Section 3.2 and Appendix D] The main text says polarity is labeled 'for frequent outcomes, e.g., frequency ≥3', but Appendix D presents this as a fixed design decision; please state clearly in the main text that the reported triangulation uses this threshold.
  4. [Fig. 2] The figure is cited in Section 4.3 but does not appear in the provided text; ensure the final version includes the figure with labeled axes, a legend, and a panel or annotation showing the number of contributing disaster×source cells over time.
  5. [Appendix E] Several citations in Appendix E (e.g., composed image/video retrieval works) are only loosely connected to the presented experiments; consider trimming these references or moving them to a clearly separated future-directions discussion to keep the related-work focus.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: extraction is evaluated on independent expert labels and triangulation is a closed-form descriptive aggregation.

full rationale

The paper's derivation chain is self-contained. The headline extraction numbers are measured against 100 expert-annotated reports (Section 3; Appendix C), and the LoRA student is trained on teacher outputs from a disjoint split with the 100 evaluation reports explicitly excluded (Section 3.1; Appendix F), so the 94.15% weighted-F1 result is not fitted to the evaluation labels. The triangulation score LoE is computed by explicit closed-form equations (2)-(5) from extracted records: strength-weighted cell counts, Laplace smoothing, equal cell averaging, and a normalized maximum-probability transform. No equation inverts or is fitted to the reported convergence; LoE=0.865 is a deterministic function of the cell-averaged PP=0.91, not a prediction validated against a target. The only self-citation (Zhang et al., 2026c) appears in Related Work as context and is not load-bearing. The acknowledged limitations about annotation-set size and sensitivity to polarity normalization, weighting, and smoothing are robustness and validity risks rather than circularity: Appendix D's decision to label only frequent outcomes is an explicit preprocessing choice, and its potential effect on the headline LoE is a measurement concern, not a self-referential derivation. Overall, the extraction claims are externally grounded and the aggregation is a transparent summary statistic.

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

The extraction F1 claims rest on the reliability of 100 expert-annotated reports and on hand-set evaluation thresholds; the triangulation LoE=0.865 rests on hand-set smoothing and strength weights, manual polarity labels for frequent outcomes, and the filtered disaster and source corpus. None of these are derived from first principles or externally benchmarked, so the ledger records them as assumptions and free parameters rather than established facts.

free parameters (5)
  • Laplace smoothing alpha = 0.1
    Hand-set in Eq. (3); controls the shrinkage of cell probabilities and therefore the LoE score; no sensitivity analysis is reported.
  • Strength weights w(sigma) = WEAK=0, MOD=1, STRONG=2
    Hand-set ordinal weights in Eq. (2); directly determine the cell counts C(c)_r and hence the global probabilities and LoE.
  • Weighted-F1 component weights = 0.30, 0.30, 0.20, 0.20
    Hand-set aggregation weights in Table 2; changing them changes the headline weighted F1 and the model ranking.
  • BERTScore threshold = 0.8
    Threshold for semantic set matching in evaluation; affects which intervention and outcome mentions are counted as matched and all downstream F1 numbers.
  • Polarity labeling frequency threshold = frequency >= 3
    Appendix D; outcome terms below the threshold are not polarity-normalized and are largely excluded from the triangulation case study (frequent terms cover 69.27% of records).
assumptions (5)
  • domain assumption The 100-report, 220-relation expert-annotated set is an accurate and consistent ground truth for intervention-outcome relations.
    Section 3 Evaluation and Appendix C; no inter-annotator agreement is reported, and the paper's Limitations note sensitivity to sampling and annotation guidelines.
  • domain assumption LLM-extracted direction and strength labels correspond to causal evidence in the reports.
    Eq. (1) defines causal records but no causal verification is performed; Limitations clarify that strength reflects reported evidence strength, not effect magnitude.
  • domain assumption The filtered corpus (four disaster types, five source types, English-only, keyword 'cash assistance') is representative enough for cross-context triangulation.
    Section 3 Data and Appendix A; the inclusion criteria may bias the cell-level evidence distributions and the resulting convergence score.
  • ad hoc to paper Laplace smoothing with alpha=0.1 and equal cell weighting is a valid aggregation for the Level-of-Evidence score.
    Eqs. (2)-(5); these choices are hand-set and no sensitivity analysis is provided.
  • domain assumption Manual polarity labels for outcome terms with frequency >=3 correctly map outcome framing to direction.
    Section 3.2 and Appendix D; labeling covers 69.27% of records and the mapping is subjective for a long tail of outcomes.
invented entities (1)
  • Level-of-Evidence (LoE) score
    purpose: Quantifies cross-context convergence as the normalized excess of the dominant cell-averaged relation probability over the 1/3 baseline.
    Defined in Eq. (5); no external calibration against established evidence grades or real intervention-effectiveness data, and the value depends on hand-set smoothing, strength weights, and cell definitions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Causal Evidence Extraction and Triangulation in Crisis Reports using Large Language Models: A ReliefWeb-based Study." pith.science (2026). https://pith.science/paper/XK4YWFZR

@misc{pith2026260804576,
  author       = {Pith},
  title        = {Pith review of: Causal Evidence Extraction and Triangulation in Crisis Reports using Large Language Models: A ReliefWeb-based Study},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XK4YWFZR}},
  note         = {Machine review of arXiv:2608.04576}
}
abstract

Humanitarian reports are long, noisy, and multi-topic, making it difficult to consolidate decision-relevant causal evidence. We present a ReliefWeb study (2000-2024) and a two-stage Large Language Model (LLM) pipeline that extracts structured intervention-outcome records with direction and strength attributes. Query-conditioned extraction restricts output to a specified intervention class, reducing retrieval-induced over-extraction, while snippet grounding links each relation to supporting text for auditability and classification. In an expert-annotated dataset of 100 reports, the best closed-source LLM achieved a weighted F1 score of 90.73% with strong cost-efficiency, while Llama-3.1-8B with supervised fine-tuning reached 94.15% weighted F1 score. We further propose context-preserving triangulation that aggregates strength-weighted evidence within disaster$\times$source cells, applies Laplace smoothing and equally weights cells to quantify cross-context convergence via a Level-of-Evidence score. Applied to cash assistance, food-related outcomes show strong positive convergence (LoE=0.865) and stable long-horizon trajectories.

Figures

Figures reproduced from arXiv: 2608.04576 by the authors.

Figure 1
Figure 1. Overall framework for causal evidence extraction and context-preserving triangulation. ReliefWeb reports [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Cumulative temporal triangulation for polarity-normalized food-related outcomes: cell￾averaged PP , PZ, PN and the number of contributing disaster×source cells over time. (prices in Appendix Tab. 5). Our pipeline also provides supervision for distillation: LoRA fine￾tuning Llama-3.1-8B-Instruct improves weighted F1 from 75.12% to 94.15% (Appendix F). 4.3 Triangulation: cross-context convergence Using Qwen-Plus (two-… view at source ↗
Figure 3
Figure 3. Annual number of ReliefWeb reports by source type. [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Frequency-based word cloud of extracted outcomes (used to motivate the food-related case study). [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 10 canonical work pages

  1. [6]

    Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, and Mingwei Chang

    The llama 3 herd of models.Preprint, arXiv:2407.21783. Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, and Mingwei Chang

    arXiv

  2. [8]

    Todd D Jick

    Gpt-4o system card.Preprint, arXiv:2410.21276. Todd D Jick

    arXiv

  3. [9]

    Faithful-First Reasoning, Planning, and Acting for Multimodal LLMs

    Faithact: Faithfulness planning and acting in mllms. arXiv preprint arXiv:2511.08409. Zixu Li, Yupeng Hu, Zhiwei Chen, Qinlei Huang, Guozhi Qiu, Zhiheng Fu, and Meng Liu. 2026a. Retrack: Evidence-driven dual-stream directional anchor calibration network for composed video re- trieval.Proceedings of the AAAI Conference on Arti- ficial Intelligence, 40(28):...

    work page Pith review arXiv

  4. [10]

    Peiyang Liu, Ziqiang Cui, Di Liang, and Wei Ye

    Mmfinereason: Closing the multimodal reasoning gap via open data-centric methods.Preprint, arXiv:2601.21821. Peiyang Liu, Ziqiang Cui, Di Liang, and Wei Ye. 2025a. Who stole your data? a method for detecting unau- thorized rag theft.Preprint, arXiv:2510.07728. Peiyang Liu, Xi Wang, Ziqiang Cui, and Wei Ye. 2025b. Queries are not alone: Clustering text emb...

    arXiv

  5. [11]

    InPro- ceedings of the 32nd ACM International Conference on Information and Knowledge Management, CIKM ’23, page 4099–4104, New York, NY , USA

    Retrieval-based unsu- pervised noisy label detection on text data. InPro- ceedings of the 32nd ACM International Conference on Information and Knowledge Management, CIKM ’23, page 4099–4104, New York, NY , USA. Associa- tion for Computing Machinery. Zheng Liu, Hao Liang, Bozhou Li, Wentao Xiong, Chong Chen, Conghui He, Wentao Zhang, and Bin Cui. 2025c. Sy...

    arXiv

  6. [12]

    Yunbo Long, Yuhan Liu, Liming Xu, and Alexandra Brintrup

    Emo- mas: Emotion-aware multi-agent system for high- stakes edge-deployable negotiation with bayesian or- chestration.Preprint, arXiv:2604.07003. Yunbo Long, Yuhan Liu, Liming Xu, and Alexandra Brintrup

    arXiv

  7. [13]

    Kexin Ma, Ruochun Jin, Wang Haotian, Wang Xi, Huan Chen, Yuhua Tang, and Qian Wang

    Emodebt: Bayesian-optimized emo- tional intelligence for strategic agent-to-agent debt recovery.Preprint, arXiv:2503.21080. Kexin Ma, Ruochun Jin, Wang Haotian, Wang Xi, Huan Chen, Yuhua Tang, and Qian Wang

    arXiv

  8. [14]

    InFindings of the Association for Computational Linguistics: EMNLP 2024, pages 4886–4901, Miami, Florida, USA

    Context- driven index trimming: A data quality perspective to enhancing precision of RALMs. InFindings of the Association for Computational Linguistics: EMNLP 2024, pages 4886–4901, Miami, Florida, USA. Asso- ciation for Computational Linguistics. Xiaoxu Ma, Xiangbo Zhang, and Zhenyu Weng

    work page 2024

Show all 22 references
  1. [15]

    Preprint, arXiv:2601.09833

    Stable and explainable personality trait evaluation in large language models with internal activations. Preprint, arXiv:2601.09833. Potsawee Manakul, Adian Liusie, and Mark Gales

  2. [16]

    InProceedings of the 2023 Conference on Empiri- cal Methods in Natural Language Processing, pages 9004–9017, Singapore

    SelfCheckGPT: Zero-resource black-box hallucina- tion detection for generative large language models. InProceedings of the 2023 Conference on Empiri- cal Methods in Natural Language Processing, pages 9004–9017, Singapore. Association for Computa- tional Linguistics. Cheng Niu,...

    2023

  3. [17]

    https://apidoc.reliefweb.int/

    Reliefweb api documentation. https://apidoc.reliefweb.int/. Service pro- vided by UN OCHA; accessed 2025-12-20. Roberta Rocca, Nicolò Tamagnone, Selim Fekih, Xi- mena Contla, and Navid Rekabsaz

    2025

  4. [18]

    Yuan Yao, Deming Ye, Peng Li, Xu Han, Yankai Lin, Zhenghao Liu, Zhiyuan Liu, Lixin Huang, Jie Zhou, and Maosong Sun

    Stable: Effi- cient hybrid nearest neighbor search via magnitude- uniformity and cardinality-robustness.Preprint, arXiv:2604.01617. Yuan Yao, Deming Ye, Peng Li, Xu Han, Yankai Lin, Zhenghao Liu, Zhiyuan Liu, Lixin Huang, Jie Zhou, and Maosong Sun

  5. [19]

    InThe Twelfth International Conference on Learning Representa- tions, ICLR 2024, Vienna, Austria, May 7-11,

    Making retrieval-augmented language models robust to irrelevant context. InThe Twelfth International Conference on Learning Representa- tions, ICLR 2024, Vienna, Austria, May 7-11,

    2024

  6. [20]

    Qianchi Zhang, Hainan Zhang, Liang Pang, Yongxin Tong, Hongwei Zheng, and Zhiming Zheng

    OpenReview.net. Qianchi Zhang, Hainan Zhang, Liang Pang, Yongxin Tong, Hongwei Zheng, and Zhiming Zheng. 2026a. Less is more: Compact clue selection for efficient retrieval-augmented generation reasoning. InPro- ceedings of the ACM Web Conference 2026, WWW ’26, page 1971–1982,...

    2026

  7. [22]

    cash assistance

    OpenRe- view.net. Wenyuan Zhang, Xinghua Zhang, Haiyang Yu, Shuaiyi Nie, Bingli Wu, Juwei Yue, Tingwen Liu, and Yong- bin Li. 2026b. Expseek: Self-triggered experience seeking for web agents.Preprint, arXiv:2601.08605. Yuanjun Zhang, Fuzel Ahamed Shaik, Suvojit Acharjee, Fahad...

    2000 arXiv

  8. [2015]

    Enhanced Response Capacity Project 2014–2015

    Operational guidance and toolkit for multipurpose cash grants. Enhanced Response Capacity Project 2014–2015. Zhiwei Chen, Yupeng Hu, Zhiheng Fu, Zixu Li, Jiale Huang, Qinlei Huang, and Yinwei Wei

    2014

  9. [2020]

    In8th International Conference on Learning Representations, ICLR 2020, Addis Ababa, Ethiopia, April 26-30,

    Bertscore: Evalu- ating text generation with BERT. In8th International Conference on Learning Representations, ICLR 2020, Addis Ababa, Ethiopia, April 26-30,

    2020

  10. [2022]

    InThe Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29,

    Lora: Low-rank adaptation of large language models. InThe Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29,

    2022

  11. [2023]

    Qwen technical report.Preprint, arXiv:2309.16609. Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-V oss, Gretchen Krueger, Tom Henighan, R...

    arXiv

  12. [2024]

    InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11,

    Self-rag: Learning to retrieve, generate, and critique through self-reflection. InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11,

    2024

  13. [2025]

    Preprint, arXiv:2412.19437

    Deepseek-v3 technical report. Preprint, arXiv:2412.19437. Jay DeYoung, Sarthak Jain, Nazneen Fatema Rajani, Eric Lehman, Caiming Xiong, Richard Socher, and Byron C Wallace

    arXiv

  14. [2026]

    Emsdialog: Synthetic multi- person emergency medical service dialogue genera- tion from electronic patient care reports via multi-llm agents.Preprint, arXiv:2604.07549. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al- Dahle, Aiesha...

    arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.