Pith. sign in

REVIEW 6 major objections 5 minor 2 cited by

Enhancing RAG with Active Learning on Conversation Records: Reject Incapables and Answer Capables

T0 review · 6 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read AL4RAG claims that a diversity-based active learner using retrieval-augmented similarity (ras) selects RAG conversation records whose annotation yields models that refuse hallucination-prone queries and answer the rest more steadily.

desk verdict A useful ras similarity variant that improves diversity-based active learning for RAG, but the undefined Rejection Rate and over-claimed 'first' framing need fixing before the headline numbers can be trusted. read the letter →

arxiv 2502.09073 v1 pith:YSZHYH7P submitted 2025-02-13 cs.CL

Xuzhao Geng , Haozhao Wang , Jun Wang , Wei Liu , Ruixuan Li This is my paper · ORCID
classification cs.CL
keywords activelearningretrieval-augmentedgenerationpreferenceoptimizationDPOhallucinationrefusalsamplediversitysimilaritymetricRAGconversationrecords
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's central claim is that the annotation bottleneck for teaching retrieval-augmented generation (RAG) models when to refuse and when to answer can be attacked with active learning tailored to RAG's three-part conversation structure. It proposes AL4RAG, which selects a diverse subset of unlabeled query-reference-response records for human labeling, then converts those labels into a preference dataset and fine-tunes the model with DPO. The key to the selection is a new similarity measure, ras, that compares queries and retrieved references separately and takes the minimum with the full-prompt similarity, so that short user queries are not drowned out by long reference texts. On an expanded RAGTruth benchmark, models trained on data selected by ras reach a rejection rate of 22.65 with 12.5% of the data, higher than the 18.66 of the strongest IDDS baseline, and also improve answer stability (ROUGE-L 22.23 vs 20.96). A sympathetic reading is that budget-limited preference-data creation for RAG can be materially improved by better sample-distance measurement.

What carries the argument

The load-bearing object is retrieval-augmented similarity (ras), defined for two samples $x$ and $y$ as $\operatorname{ras}(x,y) = \min(\cos(p_x,p_y), \tfrac{1}{2}(\cos(q_x,q_y)+\cos(r_x,r_y)))$, where $q$ is the user query, $r$ is the concatenated retrieved reference, and $p$ is the prompt combining $q$ and $r$. The min with the prompt similarity prevents long references from inflating apparent similarity, while the separate query-reference term keeps the user's intent influential. This similarity feeds an IDDS-style scoring loop that prefers samples far from the already-selected set and close to the remaining unlabeled pool, producing the subset that is annotated and turned into DPO preferences.

What would settle it

Re-run the 12.5%-budget comparison with Rejection Rate defined and measured against human judgments of whether the model should have refused, using independently re-annotated hallucination labels; if ras-selected DPO no longer beats the IDDS baseline, the reported advantage rests on the label and refusal pipeline rather than on the similarity measure.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that diversity-based active learning for RAG fails when sample similarity is computed on the user query alone or on the whole prompt, because retrieved references are long and dominate the distance; a similarity that separates the query and reference fields, averages those similarities, and caps the result by the prompt similarity selects more informative records. Given the same annotation budget, DPO training on the ras-selected preference set outperforms random, entropy, coreset, BLEU-variance, and three IDDS variants on both refusal of hallucination-prone queries and stability of correct answers, and a quarter of the ras-selected data beats full-data DPO on rejection and stability. The paper further claims that this is the first active-learning framework for RAG and the first human preference dataset built for RAG-style refusal and answer decisions.

Load-bearing premise

The comparison assumes that the RAGTruth hallucination labels, converted into preference pairs through a fixed refusal template, are correct targets for DPO, and that the unstated Rejection Rate metric measures what the model learned about when to refuse.

Editorial extensions

If this is right

  • With 12.5% of records selected by ras, DPO reaches a rejection rate of 22.65 and ROUGE-L of 22.23, beating the best IDDS variant (18.66 and 20.96) and full-data DPO, so annotation budgets can be cut to a small fraction of the available pool.
  • Selecting samples before annotation is enough; all selected records are labeled at the end, so annotators do not wait between rounds.
  • The refusal preference is constructed from single-answer records by pairing the original answer with a fixed refusal template according to hallucination labels, making single-response RAG logs usable for DPO.
  • Ablation results show that dropping the query term or adding the answer term hurts rejection or stability, identifying which part of the record the diversity score must preserve.
  • TF-IDF vectorization outperforms Sentence-BERT and a 1.5B embedding model for this selection, so surface-form diversity rather than deep semantics drives the reported gains.

Reading between the lines

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

  • The min-of-prompt and mean-of-parts construction suggests a general principle for structured records: when fields have very different lengths, aggregate field-level similarities and guard against dominance by a single long field; the same trick could transfer to other multi-field inputs such as code-and-doc or table-and-question pairs.
  • The observed decline of ras's advantage as the budget grows hints that diversity selection is most valuable when the annotation budget is tight; one testable extension is an adaptive budget rule that switches from diversity sampling to another criterion once the marginal gain flattens.
  • Because the preference set teaches refusal through a fixed template, a natural next step not pursued in the paper is measuring whether the learned refusal generalizes to new phrasing of hallucination-prone questions or to out-of-distribution queries.
  • Implicit in the paper is the idea that hallucination labels themselves are the supervision; if those labels are noisy, the same pipeline could be re-run with labels from a different source to see whether ras still selects better records.
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

6 major / 5 minor

Summary. The paper proposes AL4RAG, an active learning method for curating RAG conversation records for DPO fine-tuning. It introduces retrieval-augmented similarity (ras) (Eq. 5), which measures sample distances via separate query/reference/answer attributes and a minimum operation, and selects diverse samples under an annotation budget. The selected samples are converted into preference pairs using hallucination labels and a fixed refusal response, and the model is fine-tuned with DPO. Experiments on a RAGTruth-derived dataset report consistent gains in Rejection Rate and stability metrics over random, entropy, coreset, BLEUVar, and IDDS baselines.

Significance. The paper addresses a real bottleneck: annotation cost for building preference data in RAG. The ras distance is a sensible idea and the ablations in Table 3 support the role of query/reference separation. The direction of the reported gains is consistent across Tables 1 and 2, and the authors provide implementation details and explicit hyperparameter values. However, the evaluation is compromised by an undefined central metric and by a preference-construction pipeline that is described as human annotation but implemented with externally supplied hallucination labels and a fixed template. These issues must be resolved before the empirical claims can be interpreted. If resolved and the gains survive significance testing, the contribution would be a useful practical tool for budget-limited RAG alignment.

major comments (6)
  1. [§6.1, Rejection Rate metric] The central rejection metric, Rejection Rate (RR), is never defined. The paper reports RR in Table 1 and Figure 2 without specifying the formula, matching rule, or threshold; if RR counts exact string matches to the fixed refusal template used to construct DPO targets in §4.2, the AL4RAGras gains could largely reflect how strongly each selected subset teaches the model to copy that template. Please provide the exact definition and also report the false-refusal rate on answerable queries.
  2. [§4.2 vs. §6.1] The preference dataset is not human preference data as described. §4.2 says annotators assess hallucinations and generate label h, but §6.1 says hallucination labels were taken from RAGTruth and mapped to preferred/non-preferred answers. Moreover, every h=1 sample uses the same fixed refusal string as chosen and every h=0 sample uses the same string as rejected, so the preference set is constructed by a fixed template rather than by paired human judgments. The contributions claiming a human preference dataset, including the 'first human preference dataset tailored to the RAG scenario', are therefore not supported as written. Please clarify which labels came from humans and which were derived, and report the annotation details.
  3. [Tables 1 and 2] Results are reported as means over five runs with no standard deviations or significance tests, so the claimed 'consistent outperformance' cannot be separated from run-to-run noise. This matters because stability differences are small; for example, Table 2 shows several baselines within 20-21 ROUGE-L at 12.5%, yet no variance information is given. Please provide per-run values, error bars, and a paired significance test across the five runs.
  4. [Eq. (5) and Eq. (3)] The ras similarity uses an unmotivated 'min' over prompt similarity and the average of query/reference similarities, and the IDDS score depends on λ in Eq. (3), with neither subject to a sensitivity analysis. Since the core claim is that ras improves diversity selection, please provide ablations on the fusion function (min vs. max vs. average) and a λ sweep; otherwise the observed gains could be due to the specific hyperparameter choice rather than the principle of independent attribute measurement.
  5. [§6.1, Model paragraph] The DPO experiments use an SFT model 'fine-tuned on the training set', i.e., on the full data, while the active learning budget is applied only to the DPO preference pairs. This makes the comparison an evaluation of data selection for DPO given a fully supervised SFT stage, not the annotation-budget scenario described in the abstract and introduction. Please state whether the SFT model is trained on all data or only on the AL-selected subset, and if the former, discuss how this affects the budget interpretation.
  6. [References / novelty] The reference list includes Xu et al. (2024b), 'ActiveRAG: Revealing the Treasures of Knowledge via Active Learning', but the text never cites or discusses it. Because the abstract and §1 claim the first AL framework for RAG, this omission is load-bearing for the novelty claim. Please discuss this work explicitly and position the contribution relative to it.
minor comments (5)
  1. [§4.1, step (5)] 'top %k samples' should read 'top k% samples'.
  2. [§5.1] The word 'referneces' is a typo for 'references'.
  3. [Figure 2] The left panel's y-axis label 'Rejection' should be 'Rejection Rate' for consistency with the text and tables.
  4. [§6.1] The description of the RAGTruth expansion should state the exact filtering criteria and per-task counts; as written, it is unclear how the approximately 3,000 entries were obtained.
  5. [References] Several bibliography entries (e.g., Wang et al., 2024a,b; Xu et al., 2024a) are not cited in the text; please either cite them in context or remove them from the reference list.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: ras selection is evaluated against external RAGTruth labels and non-parametric similarity; the undefined Rejection Rate is an evaluation gap, not a constructional equivalence.

full rationale

The paper's central claim—that the ras-based AL selector (Eq. 5) identifies informative RAG conversation records—is evaluated end-to-end against an external benchmark (RAGTruth) whose hallucination labels are used to construct preference pairs, and the DPO objective (Eq. 4) is applied to those pairs plus a fixed refusal template. No parameter is fitted to a subset and then reported as a prediction of that same subset; ras is a non-parametric combination of TF-IDF cosine similarities and is compared against query-only, prompt-only, and IDDS baselines. The only concerns are verification-oriented rather than circular: 'Rejection Rate' is never defined in §6.1, and §4.2's described human-annotation procedure is not what §6.1 implements, since existing RAGTruth labels are used. Neither concern makes the derivation reduce to its own inputs, so the circularity score is 0.

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

The method introduces no new physical entities. Its load-bearing assumptions are the validity of existing hallucination labels, the binary preference mapping with a fixed refusal template, the diversity objective, and the choice of TF-IDF. The main free design choices are the IDDS lambda, the ras min and average combination, the DPO beta, and the AL schedule.

free parameters (4)
  • lambda in IDDS score = not reported
    Eq. 3 blends similarity to the unselected pool with dissimilarity to the selected set; the paper gives no value or tuning procedure.
  • ras combination weights and min operation = 1/2 average and min, no alternatives tested
    Eq. 5 fixes the average of query and reference similarity and the min with prompt similarity by design; no theoretical justification or sensitivity analysis is provided.
  • DPO beta = not reported
    Eq. 4 uses beta to control alignment speed; the paper does not report its value or tuning.
  • AL iteration rounds and initial random set size = 5 rounds; k unspecified
    Section 6.1 sets iteration rounds to five, but the initial randomly selected set size and per-round batch size are not specified.
assumptions (4)
  • domain assumption RAGTruth hallucination labels are accurate ground-truth labels for the Llama-2-7B-chat responses used.
    Section 6.1 says labels were identified from RAGTruth and assigned preferred and rejected answers; no re-annotation or label quality check is reported.
  • domain assumption A fixed refusal template is the correct target for all hallucination-prone samples, and original responses are correct targets for all non-hallucinated samples.
    Section 4.2 constructs the preference set this way; if some hallucinated queries should be answered or some correct answers are better phrased, the DPO targets are mis-specified.
  • domain assumption Diversity-based IDDS selection is an appropriate proxy for annotation value in the RAG setting.
    Section 3.2 adopts IDDS from Tsvigun et al. without evidence that maximizing diversity of TF-IDF vectors correlates with downstream DPO gain.
  • domain assumption TF-IDF cosine similarity captures task-relevant differences better than semantic embeddings.
    Section 6.4 reports TF-IDF outperforms Sentence-BERT and stella on one dataset, with no theoretical reason or further validation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing RAG with Active Learning on Conversation Records: Reject Incapables and Answer Capables." pith.science (2026). https://pith.science/paper/YSZHYH7P

@misc{pith2026250209073,
  author       = {Pith},
  title        = {Pith review of: Enhancing RAG with Active Learning on Conversation Records: Reject Incapables and Answer Capables},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YSZHYH7P}},
  note         = {Machine review of arXiv:2502.09073}
}
read the original abstract

Retrieval-augmented generation (RAG) is a key technique for leveraging external knowledge and reducing hallucinations in large language models (LLMs). However, RAG still struggles to fully prevent hallucinated responses. To address this, it is essential to identify samples prone to hallucination or guide LLMs toward correct responses, which experts then annotate to develop high-quality datasets for refining LLMs. However, the growing scarcity of such datasets makes their creation challenging. This paper proposes using the vast amount of conversations from widespread LLM usage to build these datasets, training LLMs to avoid hallucination-prone questions while accurately responding to manageable ones. Given the impracticality of expert-annotating all conversation records, the paper introduces AL4RAG, which uses active learning to select the most suitable conversation samples for annotation, optimizing performance within an annotation budget. Additionally, recognizing that traditional active learning methods are not fully compatible with RAG due to unsuitable distance metrics, we develop a novel sample distance measurement for RAG active learning. Extensive experiments show that our method consistently outperforms baselines across multiple metrics.

Figures

Figures reproduced from arXiv: 2502.09073 by the authors.

Figure 1
Figure 1. (a) An example regarding our tasks. In the scenario on the left, the original model provides an incorrect response to the user; in this case, we expect the model to decline answering the question, whereas in the scenario on the right, where the original model delivers a correct response, we aim for it to generate accurate responses more consistently. (b) Overall framework of our approach. The example of preference s… view at source ↗
Figure 2
Figure 2. Part of the experimental results of query similarity, prompt similarity and ras, which shows the impact of different similarity measurement on model performance. The left graph shows the performance of handling hallucination-prone queries, and the graph on the right shows the performance of handling model-answerable queries. Further detailed results are provided in Section 6.2. This highlights the robustness and ada… view at source ↗
Figure 3
Figure 3. The performance of the original model, the full-data SFT model, and the full-data DPO-trained model, with the performance of the model trained via DPO with 25% data selected by our method as comparison. (a) Rejection rate of rejection performance; (b) Rouge-L of stability performance; (c) BERTScore of stability performance [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. AI-Driven Climate Policy Scenario Generation for Sub-Saharan Africa

    cs.AI 2025-05 conditional novelty 4.0 of 10

    A RAG pipeline using llama3.2-3B and UN COP documents generated 34 policy scenarios for Sub-Saharan Africa, 30 passed author validation, but automated evaluation showed mixed agreement with human judgment.

  2. A Survey of the State-of-the-Art in Conversational Question Answering Systems

    cs.CL 2025-09 conditional novelty 2.0 of 10

    A review that categorizes ConvQA components, techniques, models, and datasets, with no new experimental result.

Reference graph

Works this paper leans on

34 extracted references · 17 canonical work pages · cited by 2 Pith papers

  1. [1]

    Generative ai text clas- sification using ensemble llm approaches

    Abburi, H., Suesserman, M., Pudota, N., Veeramani, B., Bowen, E., and Bhattacharya, S. Generative ai text clas- sification using ensemble llm approaches. arXiv preprint arXiv:2309.07755,

    arXiv

  2. [3]

    Contextual diversity for active learning

    Agarwal, S., Arora, H., Anand, S., and Arora, C. Contextual diversity for active learning. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XVI 16 , pp. 137–153. Springer,

    work page 2020

  3. [6]

    and Tsoumakas, G

    Gidiotis, A. and Tsoumakas, G. Should we trust this sum- mary? bayesian abstractive summarization to the rescue. In Findings of the Association for Computational Linguis- tics: ACL 2022, pp. 4119–4131,

    work page 2022

  4. [8]

    A comprehensive survey on process-oriented automatic text summarization with exploration of llm-based methods

    Jin, H., Zhang, Y ., Meng, D., Wang, J., and Tan, J. A comprehensive survey on process-oriented automatic text summarization with exploration of llm-based methods. arXiv preprint arXiv:2403.02901,

    arXiv

  5. [11]

    Unleashing the Power of Continual Learning on Non-Centralized Devices: A Survey

    Li, Y ., Wang, H., Xu, W., Xiao, T., Liu, H., Tu, M., Wang, Y ., Yang, X., Zhang, R., Yu, S., Guo, S., and Li, R. Unleash- ing the power of continual learning on non-centralized devices: A survey. CoRR, abs/2412.13840, 2024b. Li, Z., Qu, L., Cohen, P. R., Tumuluri, R., and Haffari, G. The best of both worlds: Combining human and machine translations for m...

    work page Pith review arXiv

  6. [12]

    Low-resource interactive active labeling for fine-tuning language models

    Maekawa, S., Zhang, D., Kim, H., Rahman, S., and Hr- uschka, E. Low-resource interactive active labeling for fine-tuning language models. In Findings of the Associ- ation for Computational Linguistics: EMNLP 2022, pp. 3230–3242,

    work page 2022

  7. [13]

    Active learning principles for in-context learning with large language models

    Margatina, K., Schick, T., Aletras, N., and Dwivedi-Yu, J. Active learning principles for in-context learning with large language models. In Findings of the Association for Computational Linguistics: EMNLP 2023, pp. 5011– 5034,

    work page 2023

  8. [14]

    Mteb: Massive text embedding benchmark

    Muennighoff, N., Tazi, N., Magne, L., and Reimers, N. Mteb: Massive text embedding benchmark. arXiv preprint arXiv:2210.07316,

    arXiv

Show all 34 references
  1. [15]

    Ragtruth: A hallucination corpus for developing trustworthy retrieval-augmented language models

    Niu, C., Wu, Y ., Zhu, J., Xu, S., Shum, K., Zhong, R., Song, J., and Zhang, T. Ragtruth: A hallucination corpus for developing trustworthy retrieval-augmented language models. arXiv preprint arXiv:2401.00396,

    arXiv

  2. [16]

    Sentence-bert: Sentence embeddings using siamese bert-networks

    Reimers, N. Sentence-bert: Sentence embeddings using siamese bert-networks. arXiv preprint arXiv:1908.10084,

    1908 arXiv

  3. [17]

    Revisiting uncertainty-based query strategies for active learning with transformers

    Schr¨oder, C., Niekler, A., and Potthast, M. Revisiting uncertainty-based query strategies for active learning with transformers. arXiv preprint arXiv:2107.05687,

    arXiv

  4. [18]

    and Savarese, S

    Sener, O. and Savarese, S. Active learning for convolutional neural networks: A core-set approach. arXiv preprint arXiv:1708.00489,

    arXiv

  5. [20]

    Large language models for data annotation: A survey

    10 Enhancing RAG with Active Learning on Conversation Records Tan, Z., Li, D., Wang, S., Beigi, A., Jiang, B., Bhattacharjee, A., Karami, M., Li, J., Cheng, L., and Liu, H. Large language models for data annotation: A survey. arXiv preprint arXiv:2402.13446,

    arXiv

  6. [21]

    and Goel, A

    Taneja, K. and Goel, A. Can active label correction im- prove llm-based modular ai systems? In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pp. 9019–9031,

    2024

  7. [22]

    I., Burnell, R., Bai, L., Gulati, A., Tanzer, G., Vincent, D., Pan, Z., Wang, S., et al

    Team, G., Georgiev, P., Lei, V . I., Burnell, R., Bai, L., Gulati, A., Tanzer, G., Vincent, D., Pan, Z., Wang, S., et al. Gemini 1.5: Unlocking multimodal understand- ing across millions of tokens of context. arXiv preprint arXiv:2403.05530,

    arXiv

  8. [23]

    Llama 2: Open foundation and fine- tuned chat models

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

    arXiv

  9. [24]

    Active learning for abstractive text summarization

    Tsvigun, A., Lysenko, I., Sedashov, D., Lazichny, I., Damirov, E., Karlov, V ., Belousov, A., Sanochkin, L., Panov, M., Panchenko, A., et al. Active learning for abstractive text summarization. In Findings of the Associ- ation for Computational Linguistics: EMNLP 2022, pp. 5128–5152,

    2022

  10. [25]

    A survey of active learning algorithms for super- vised remote sensing image classification

    Tuia, D., V olpi, M., Copa, L., Kanevski, M., and Munoz- Mari, J. A survey of active learning algorithms for super- vised remote sensing image classification. arXiv preprint arXiv:2104.07784,

    arXiv

  11. [26]

    and Shang, Y

    Wang, D. and Shang, Y . A new active labeling method for deep learning. In 2014 International joint conference on neural networks (IJCNN), pp. 112–119. IEEE,

    2014

  12. [28]

    A comprehensive survey on deep active learning and its applications in medical image analysis

    Wang, H., Jin, Q., Li, S., Liu, S., Wang, M., and Song, Z. A comprehensive survey on deep active learning and its applications in medical image analysis. arXiv preprint arXiv:2310.14230,

    arXiv

  13. [29]

    Feddse: Distribution-aware sub- model extraction for federated learning over resource- constrained devices

    Wang, H., Jia, Y ., Zhang, M., Hu, Q., Ren, H., Sun, P., Wen, Y ., and Zhang, T. Feddse: Distribution-aware sub- model extraction for federated learning over resource- constrained devices. In Proceedings of the ACM on Web Conference 2024, WWW 2024, Singapore, May 13-17, 2024, ...

    2024 arXiv

  14. [30]

    Freeal: Towards human-free active learning in the era of large language models

    Xiao, R., Dong, Y ., Zhao, J., Wu, R., Lin, M., Chen, G., and Wang, H. Freeal: Towards human-free active learning in the era of large language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pp. 14520–14535,

    2023

  15. [31]

    Z., Gomez, A

    Xiao, T. Z., Gomez, A. N., and Gal, Y . Wat zei je? de- tecting out-of-distribution translations with variational transformers. arXiv preprint arXiv:2006.08344,

    2006 arXiv

  16. [32]

    Deploy- ing foundation model powered agent services: A survey

    Xu, W., Chen, J., Zheng, P., Yi, X., Tian, T., Zhu, W., Wan, Q., Wang, H., Fan, Y ., Su, Q., and Shen, X. Deploy- ing foundation model powered agent services: A survey. CoRR, abs/2412.13437, 2024a. Xu, Z., Liu, Z., Liu, Y ., Xiong, C., Yan, Y ., Wang, S., Yu, S., Liu, Z., and ...

    arXiv

  17. [33]

    Llmaaa: Making large language models as active annotators

    Zhang, R., Li, Y ., Ma, Y ., Zhou, M., and Zou, L. Llmaaa: Making large language models as active annotators. In Findings of the Association for Computational Linguis- tics: EMNLP 2023, pp. 13088–13103,

    2023

  18. [34]

    A survey of active learning for natural language processing

    Zhang, Z., Strubell, E., and Hovy, E. A survey of active learning for natural language processing. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pp. 6166–6190,

    2022

  19. [1995]

    Diversity- aware batch active learning for dependency parsing

    Shi, T., Benton, A., Malioutov, I., and˙Irsoy, O. Diversity- aware batch active learning for dependency parsing. In Proceedings of the 2021 Conference of the North Amer- ican Chapter of the Association for Computational Lin- guistics: Human Language Technologies, pp. 2616–2626,

    2021

  20. [2014]

    Fed- cda: Federated learning with cross-rounds divergence- aware aggregation

    Wang, H., Xu, H., Li, Y ., Xu, Y ., Li, R., and Zhang, T. Fed- cda: Federated learning with cross-rounds divergence- aware aggregation. In The Twelfth International Confer- ence on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11,

    2024

  21. [2018]

    J., Shen, Y ., Wallis, P., Allen-Zhu, Z., Li, Y ., Wang, S., Wang, L., and Chen, W

    Hu, E. J., Shen, Y ., Wallis, P., Allen-Zhu, Z., Li, Y ., Wang, S., Wang, L., and Chen, W. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685,

    arXiv

  22. [2020]

    Active learning for abstractive text summa- rization via llm-determined curriculum and certainty gain maximization

    Li, D., Zhang, Y ., Wang, Z., Tan, S., Kosugi, S., and Oku- mura, M. Active learning for abstractive text summa- rization via llm-determined curriculum and certainty gain maximization. In Findings of the Association for Com- putational Linguistics: EMNLP 2024 , pp. 8959–8971, ...

    2024

  23. [2021]

    Rs-dpo: A hybrid rejection sampling and direct prefer- ence optimization method for alignment of large language models

    Khaki, S., Li, J., Ma, L., Yang, L., and Ramachandra, P. Rs-dpo: A hybrid rejection sampling and direct prefer- ence optimization method for alignment of large language models. arXiv preprint arXiv:2402.10038,

    arXiv

  24. [2022]

    and Reuter, C

    Bayer, M. and Reuter, C. Activellm: Large language model- based active learning for textual few-shot scenarios.arXiv preprint arXiv:2405.10808,

  25. [2023]

    L., Almeida, D., Altenschmidt, J., Altman, S., Anadkat, S., et al

    Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F. L., Almeida, D., Altenschmidt, J., Altman, S., Anadkat, S., et al. Gpt-4 technical report. arXiv preprint arXiv:2303.08774,

    arXiv

  26. [2024]

    Training a helpful and harmless assistant with rein- forcement learning from human feedback

    Bai, Y ., Jones, A., Ndousse, K., Askell, A., Chen, A., Das- Sarma, N., Drain, D., Fort, S., Ganguli, D., Henighan, T., et al. Training a helpful and harmless assistant with rein- forcement learning from human feedback. arXiv preprint arXiv:2204.05862,

    arXiv

Pith tools

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