Pith. sign in

REVIEW 4 major objections 5 minor 2 cited by

ToxiLab: How Well Do Open-Source LLMs Generate Synthetic Toxicity Data?

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Supervised fine-tuning lets open-source LLMs generate toxic data that approaches GPT-4 for training detectors.

desk verdict Fine-tuning open-source LLMs for toxic data synthesis gives real downstream gains, but the data-quality claims are under-evidenced because synthetic positives are added to training without quality validation. read the letter →

arxiv 2411.15175 v4 pith:GTGGIJXG submitted 2024-11-18 cs.CL cs.AI

Zheng Hui , Zhaoxiao Guo , Hang Zhao , Juanyong Duan , Lin Ai , Yinheng Li , Julia Hirschberg , Congrui Huang This is my paper · ORCID
classification cs.CLcs.AI
keywords syntheticdatagenerationtoxiccontentdetectionhatespeechopen-sourceLLMssupervisedfine-tuningLoRAaugmentationmoderation
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

ToxiLab asks whether open-source language models can produce synthetic toxic text good enough to train content-moderation classifiers. The paper argues that prompt engineering alone is too weak because safety alignments make models refuse or repeat, but supervised fine-tuning with LoRA unlocks reliable and diverse generation. Across six open-source models and five toxicity datasets, fine-tuned Mistral consistently outperforms the other open models and approaches the GPT-4 baseline on downstream F1 and accuracy. If this holds, content moderators can augment detection datasets cheaply and transparently without relying on proprietary APIs.

What carries the argument

The mechanism is a two-stage synthesis pipeline: controlled prompting with a template containing role, requirement, and few-shot examples, followed by LoRA (Low-Rank Adaptation, a parameter-efficient fine-tuning method) on the target toxic-content dataset, then generation, deduplication, and mixing of all outputs into the original training set. The load-bearing object is the fine-tuned Mistral model plus a data-mixing strategy that blends hate, sex, and violence categories, and optionally GPT positives, so the generator produces varied and realistic toxic examples rather than refusals or repetitions. Algorithm 1 makes the augmentation loop explicit: fine-tune the generator on $D_{\text{train}}$, generate harmful samples per prompt template, combine them with $D_{\text{train}}$, and train the downstream MLP classifier on the augmented set.

What would settle it

Annotate a random sample of the synthetic harmful examples the pipeline would add to the training set: if independent annotators judge a large fraction (say, more than 20%) non-toxic or ambiguous, or if removing unverified samples raises downstream F1, the claim that generated positives are reliable fails.

Watch

Extended reading notes

Core claim

On the paper's own account, the central discovery is that supervised fine-tuning of open-source LLMs, particularly Mistral-7B, yields synthetic harmful data that is diverse and reliable enough to improve a downstream toxicity classifier, approaching the performance of GPT-4-generated data. The authors compare six open models in a prompt-engineering stage and find Mistral and Vicuna highest in success rate and human-evaluated quality, while safety-aligned models like Gemma reject harmful prompts. They then fine-tune Mistral and Vicuna with LoRA on proprietary binary datasets covering hate speech, sexual content, violence, self-harm, and political content. Fine-tuned models show higher F1 and accuracy than prompt-only generation on several datasets, and mixing hate, sex, and violence categories during fine-tuning improves generalization; mixes that add GPT-generated positives help most on self-harm. The intended conclusion is that fine-tuned open-source LLMs offer a scalable, cost-effective alternative to proprietary GPT-based data synthesis for content moderation, provided overfitting and duplication are controlled.

Load-bearing premise

The load-bearing premise is that every synthetic sample generated in response to a harmful-data prompt is genuinely harmful, so adding those samples as positive training examples improves rather than corrupts the downstream detector.

Editorial extensions

If this is right

  • A fine-tuned Mistral generator can serve as a practical substitute for GPT-4 in toxic-data synthesis, with downstream F1 and accuracy close to or above the GPT-4 baseline on several datasets and at lower cost.
  • Because prompt-only generation is blocked by safety alignment and yields repetitive, formulaic output, a serious open-source toxicity-synthesis pipeline should include supervised fine-tuning rather than relying on prompt engineering alone.
  • Fine-tuning on a mixture of toxicity categories (hate, sexual content, violence) improves downstream generalization compared with single-category fine-tuning, and mixing in GPT-generated positives gives the largest gain on self-harm detection.
  • Fine-tuning epochs must be calibrated: performance plateaus around epoch 3 and degrades by epoch 5 as the generator repeats and memorizes.
  • Deployed in real-world content moderation, the fine-tuned models reduce manual intervention for real-time filtering while requiring monitoring and periodic retraining to keep up with evolving harmful language.

Reading between the lines

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

  • The paper leaves implicit that the fine-tuning gain may be partly distribution matching: the generator learns the target dataset's style, so synthetic positives align closely with the classifier's training distribution. Fine-tuning on one toxicity domain and evaluating augmentation on another would test this.
  • The paper treats every output from a harmful-data prompt as a positive training example. A testable extension is to human-verify or confidence-filter those generated positives before adding them to $D_{\text{train}}$ and compare downstream F1.
  • If the generator is deployed broadly, its outputs could propagate biases of the proprietary training data into moderation classifiers; auditing synthetic positives for demographic disparities is a natural follow-up.
  • The 'approaches GPT-4' conclusion is drawn with an MLP downstream detector; using a stronger classifier such as a fine-tuned transformer could shrink or reverse the apparent gain from synthetic data.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper presents a two-stage pipeline for generating synthetic toxic content with open-source LLMs: first prompt engineering, then supervised fine-tuning (LoRA) on proprietary toxicity datasets. Six open-source models are compared against a GPT-4 baseline across five datasets, with downstream evaluation via a two-layer MLP classifier using F1 and accuracy. The abstract claims that Mistral outperforms other open models, that supervised fine-tuning significantly improves data reliability and diversity, and that fine-tuned open-source models approach GPT-4 performance while being more scalable and cost-effective. The paper also includes an ablation study, duplication-rate analysis, error analysis, and a discussion of deployment challenges.

Significance. If its central claims hold, the paper would provide a practical recipe for building toxic-content detection training sets without proprietary APIs, which is a timely and useful contribution for content moderation. The study is systematic in scope (six models, five datasets, both prompt-based and fine-tuned generation, plus human evaluation in stage 1), and it attempts to quantify duplication and analyze generation errors, which is more than many synthetic-data papers do. The deployment discussion is a strength in that it grounds the work in a real setting. However, the significance is conditional on resolving several load-bearing evaluation gaps: the quality/label-noise of fine-tuned generated samples is not measured, no statistical uncertainty is reported for the headline comparisons, and the fine-tuned results on the Hate dataset appear to contradict the global claim that fine-tuning improves reliability.

major comments (4)
  1. [Section 3.3, Algorithm 1; Section 5.6] The augmentation step adds every generated harmful sample to Dtrain with only duplicate removal as a filter, but Section 5.6 documents that fine-tuned models produce malformed and non-toxic outputs (e.g., random character strings, 'Sir, I would like to insult you, but I must remain polite'). Because these samples are labeled as positive when added to the training set, the magnitude of label noise is unquantified. The downstream F1/accuracy gains in Tables 3 and 4 could therefore stem from increased training-set size or distribution shift rather than from 'reliable and diverse' toxic data, which is the property the abstract claims supervised fine-tuning enhances. Please report a success rate or toxicity-precision measure for SFT-generated samples (analogous to the stage-1 success rates in Table 2), or apply a quality filter before augmentation and describe its effect.
  2. [Section 5.2, Table 4] The abstract and Section 5.2 state that supervised fine-tuning significantly improves data reliability and diversity, but Table 4 shows that on the Hate dataset all fine-tuned versions (hate_epoch1 through hate_epoch5, F1 0.540-0.590) underperform the prompt-based Mistral baseline (F1 0.608). The claim of improvement is thus not consistent across the reported results. Please either restrict the reliability/diversity claim to the datasets and configurations that actually improve (e.g., Sex and mixture settings) or provide an explanation for the Hate degradation, including whether it reflects label noise, overfitting, or a mismatch in the generated data distribution.
  3. [Tables 3 and 4] No confidence intervals, standard deviations, or significance tests are reported for the downstream F1/accuracy values, despite the statement in Section 4.2 that multiple training runs with different random seeds were performed. Several headline comparisons involve small differences (e.g., Hate: Mistral 0.608 vs. GPT-4 0.628; Mixture Epoch 5 0.672 vs. Mix_GPT 0.667), which may be within run-to-run variation. Reporting the variance across seeds or a paired significance test is necessary to support the claims that Mistral 'outperforms' other open models and that fine-tuned open models 'approach' GPT-4.
  4. [Section 4.2, Section 4.3] The downstream MLP evaluation is underspecified: the paper gives the architecture (two 600-unit hidden layers, ReLU, softmax) but never states the input feature representation (e.g., bag-of-words, TF-IDF, sentence embeddings, or token embeddings). Without this information, the F1/accuracy numbers cannot be interpreted or reproduced, and it is unclear whether the augmentation gains are specific to the chosen representation. Please specify the exact input features and, ideally, include a second downstream model (e.g., a fine-tuned transformer) to test whether the augmentation conclusions generalize beyond the MLP setting.
minor comments (5)
  1. [Section 5.2] The model name is inconsistently written as both 'Mistral' and 'Mixtral' in Section 5.2 and in the error-analysis examples; please use the actual model name consistently throughout.
  2. [Table 2] Table 2 is difficult to read: several cells contain ambiguous ranges (e.g., '≥ 65% ≤ 10%' for LLaMa-7B), and the blank cells under 'Human Eval on Quality' are not explained. Please define precisely what success rate and human-eval scores mean, add the Fleiss' kappa values mentioned in the text, and use explicit placeholders such as '—' for missing entries.
  3. [Section 3.2] The cross-entropy loss notation is under-specified: y_ij and y-hat_ij are not defined as one-hot or probability terms, and the summation indices are not fully explained. A brief definition of the variables would improve clarity.
  4. [Appendix D, Table 7] The ablation table reports precision, recall, F1, and accuracy for several configurations, but the naming is inconsistent (e.g., 'hate_v1', 'multi_hate_v1', 'hate_sex_v3') and the relationship to the main-experiment 'Mixture' rows in Table 3 is not stated. Please clarify how these configurations correspond to the mixture settings used in the main results.
  5. [Section 6] The deployment section claims successful real-world use by a 'leading cloud provider' but provides no quantitative deployment results or evaluation details. As written, this is an anecdotal claim; either add concrete metrics or mark it clearly as a qualitative report.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's synthetic-data claims are evaluated on held-out test sets and do not reduce to their inputs by construction.

full rationale

ToxiLab is an empirical study that compares synthetic toxicity data produced by prompted and fine-tuned open-source LLMs, with downstream MLP classifiers evaluated on separate 2,500-entry evaluation datasets drawn from each source corpus. The generation pipeline in Algorithm 1 fine-tunes an LLM on Dtrain, generates Dharmful, and then trains a downstream model on Dtrain ∪ Dharmful; this is a standard augmentation loop, and none of the reported F1/accuracy numbers are defined in terms of the generation success criteria or the fine-tuning loss. The success rates and human evaluation scores in Table 2 are direct measurements, not quantities forced by the construction. The paper does cite the authors' prior work ToxiCraft (Hui et al., 2024), but only as related work and as inspiration for data mixing; it is not used to prove the paper's central claims or to rule out alternatives. The most substantial validity concern—that synthetic positives added without label verification may inject label noise, as suggested by the generation errors in Section 5.6—is an external correctness or measurement-quality issue, not a circularity issue. No equation is shown to be equivalent to its input, no fitted parameter is renamed as a prediction, and no uniqueness theorem is imported from the authors' prior work. The derivation chain from generation to downstream evaluation is self-contained and the claims are empirically testable, so the appropriate circularity score is 0.

Assumptions & free parameters 2 free parameters · 2 assumptions · 0 invented entities

The central result (fine-tuning open-source LLMs improves synthetic toxic data) relies on the representativeness and correct labeling of the proprietary evaluation datasets, and on the assumption that synthetic positives are correctly labeled. The paper also implicitly assumes the MLP downstream model and its unspecified text encoding are a valid measure of data quality.

free parameters (2)
  • Fine-tuning epoch count (selected on test set) = 3 for Mixture, 5 for Mix_GPT
    Tables 3 and 4 compare epochs 1, 2, 3, and 5 and emphasize the best scores without a held-out dev set, effectively tuning the epoch hyperparameter on the test set.
  • Balanced fine-tuning dataset size = 3000 positive / 3000 negative
    Section 5.2 states all fine-tuning experiments maintain a balanced dataset of 3000 positive and 3000 negative samples; this choice affects generation quality but is not varied or justified.
assumptions (2)
  • domain assumption The five proprietary datasets (hate, sexual, violence, self-harm, political) are representative of the harmful content categories they are meant to cover, and the binary labels from multi-round human annotation are correct.
    Section 4.1 describes the datasets as sourced from online public datasets, company service collections, and proprietary sources, but no dataset is released or described in detail; the evaluation results and conclusions about 'toxic content detection' rest on this representativeness.
  • ad hoc to paper Synthetic samples generated by the fine-tuned LLM, labeled as positive (harmful) by the generation prompt, are largely correct labels and do not inject excessive label noise into the augmented training set.
    The paper's augmentation pipeline (Algorithm 1) adds all generated harmful samples to Dtrain and trains the downstream model; no human verification or annotation of the synthetic labels is reported for the fine-tuning stage, so the findings depend on this labeling assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ToxiLab: How Well Do Open-Source LLMs Generate Synthetic Toxicity Data?." pith.science (2026). https://pith.science/paper/GTGGIJXG

@misc{pith2026241115175,
  author       = {Pith},
  title        = {Pith review of: ToxiLab: How Well Do Open-Source LLMs Generate Synthetic Toxicity Data?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GTGGIJXG}},
  note         = {Machine review of arXiv:2411.15175}
}
read the original abstract

Effective toxic content detection relies heavily on high-quality and diverse data, which serve as the foundation for robust content moderation models. Synthetic data has become a common approach for training models across various NLP tasks. However, its effectiveness remains uncertain for highly subjective tasks like hate speech detection, with previous research yielding mixed results. This study explores the potential of open-source LLMs for harmful data synthesis, utilizing controlled prompting and supervised fine-tuning techniques to enhance data quality and diversity. We systematically evaluated 6 open source LLMs on 5 datasets, assessing their ability to generate diverse, high-quality harmful data while minimizing hallucination and duplication. Our results show that Mistral consistently outperforms other open models, and supervised fine-tuning significantly enhances data reliability and diversity. We further analyze the trade-offs between prompt-based vs. fine-tuned toxic data synthesis, discuss real-world deployment challenges, and highlight ethical considerations. Our findings demonstrate that fine-tuned open source LLMs provide scalable and cost-effective solutions to augment toxic content detection datasets, paving the way for more accessible and transparent content moderation tools.

Figures

Figures reproduced from arXiv: 2411.15175 by the authors.

Figure 1
Figure 1. Our experiment design is detailed: we first conducted prompt engineering on the models (stage 1), and [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Ablation Study on Data mixing Mistral and Vicuna consistently achieved the high￾est success rates and human evaluation scores among open models ( [PITH_FULL_IMAGE:figures/full_fig_p006_2.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. Lower-Resource, Higher Scores: Language Bias in LLM Evaluators

    cs.CL 2026-07 conditional novelty 7.0 of 10

    Multilingual LLM evaluators systematically inflate scores for lower-resource languages, and the standard pairwise-accuracy metric cannot detect the resulting safety-threshold disparities.

  2. TRIDENT: Benchmarking LLM Safety in Finance, Medicine, and Law

    cs.CL 2025-07 conditional novelty 5.0 of 10

    Trident-Bench provides 2,652 professionally validated harmful prompts across finance, law, and medicine, and shows that domain-specialized LLMs often comply with unethical requests more than generalist models.

Reference graph

Works this paper leans on

36 extracted references · 9 canonical work pages · cited by 2 Pith papers

  1. [1]

    Lin Ai, Tharindu Kumarage, Amrita Bhattacharjee, Zizhou Liu, Zheng Hui, Michael Davinroy, James Cook, Laura Cassani, Kirill Trapeznikov, Matthias Kirchner, Arslan Basharat, Anthony Hoogs, Joshua Garland, Huan Liu, and Julia Hirschberg. 2024. https://aclanthology.org/2024.emnlp-main.716 Defending against social engineering attacks in the age of llms . In E...

    work page 2024

  2. [2]

    Ebtesam Almazrouei, Hamza Alobeidli, Abdulaziz Alshamsi, Alessandro Cappelli, Ruxandra Cojocaru, Mérouane Debbah, Étienne Goffinet, Daniel Hesslow, Julien Launay, Quentin Malartic, Daniele Mazzotta, Badreddine Noune, Baptiste Pannier, and Guilherme Penedo. 2023. http://arxiv.org/abs/2311.16867 The falcon series of open language models

    arXiv 2023

  3. [3]

    Sai Saketh Aluru, Binny Mathew, Punyajoy Saha, and Animesh Mukherjee. 2020. http://arxiv.org/abs/2004.06465 Deep learning models for multilingual hate speech detection

    arXiv 2020

  4. [4]

    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-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litwin...

    arXiv 2020

  5. [5]

    Camilla Casula and Sara Tonelli. 2023. https://doi.org/10.18653/v1/2023.eacl-main.244 Generation-based data augmentation for offensive language detection: Is it worth it? In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 3359--3377, Dubrovnik, Croatia. Association for Computational Linguistics

    work page doi:10.18653/v1/2023.eacl-main.244 2023

  6. [6]

    Camilla Casula, Sebastiano Vecellio Salto, Alan Ramponi, and Sara Tonelli. 2024. https://doi.org/10.18653/v1/2024.emnlp-main.1099 Delving into qualitative implications of synthetic data for hate speech detection . In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 19709--19726, Miami, Florida, USA. Association...

    doi:10.18653/v1/2024.emnlp-main.1099 2024

  7. [7]

    Gonzalez, Ion Stoica, and Eric P

    Wei-Lin Chiang, Zhuohan Li, Zi Lin, Ying Sheng, Zhanghao Wu, Hao Zhang, Lianmin Zheng, Siyuan Zhuang, Yonghao Zhuang, Joseph E. Gonzalez, Ion Stoica, and Eric P. Xing. 2023. https://lmsys.org/blog/2023-03-30-vicuna/ Vicuna: An open-source chatbot impressing gpt-4 with 90\

    2023

  8. [8]

    Amanda da Silva Oliveira, Thiago de Carvalho Cecote, Jo \ a o Paulo Reis Alvarenga, Vander Luis de Souza Freitas, and Eduardo Jos \'e da Silva Luz. 2024. https://aclanthology.org/2024.propor-1.11 Toxic speech detection in P ortuguese: A comparative study of large language models . In Proceedings of the 16th International Conference on Computational Proces...

    work page 2024

Show all 36 references
  1. [9]

    Bosheng Ding, Chengwei Qin, Ruochen Zhao, Tianze Luo, Xinze Li, Guizhen Chen, Wenhan Xia, Junjie Hu, Anh Tuan Luu, and Shafiq Joty. 2024. http://arxiv.org/abs/2403.02990 Data augmentation using large language models: Data perspectives, learning paradigms and challenges

    2024 arXiv

  2. [10]

    Marzieh Fadaee, Arianna Bisazza, and Christof Monz. 2017. Data augmentation for low-resource neural machine translation. arXiv preprint arXiv:1705.00440

    2017 arXiv

  3. [11]

    Matt W Gardner and SR Dorling. 1998. Artificial neural networks (the multilayer perceptron)—a review of applications in the atmospheric sciences. Atmospheric environment, 32(14-15):2627--2636

    1998

  4. [12]

    Hamideh Ghanadian, Isar Nejadgholi, and Hussein Al Osman. 2024. Socially aware synthetic data generation for suicidal ideation detection using large language models. IEEE Access

    2024

  5. [13]

    Thomas Hartvigsen, Saadia Gabriel, Hamid Palangi, Maarten Sap, Dipankar Ray, and Ece Kamar. 2022. https://doi.org/10.18653/v1/2022.acl-long.234 T oxi G en: A large-scale machine-generated dataset for adversarial and implicit hate speech detection . In Proceedings of the 60th A...

    2022 doi

  6. [14]

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. https://openreview.net/forum?id=nZeVKeeFYf9 Lo RA : Low-rank adaptation of large language models . In International Conference on Learning Representations

    2022

  7. [15]

    Zheng Hui, Zhaoxiao Guo, Hang Zhao, Juanyong Duan, and Congrui Huang. 2024. https://doi.org/10.18653/v1/2024.findings-emnlp.970 T oxi C raft: A novel framework for synthetic generation of harmful information . In Findings of the Association for Computational Linguistics: EMNLP...

    2024 doi

  8. [16]

    Albert Q Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, et al. 2023. Mistral 7b. arXiv preprint arXiv:2310.06825

    2023 arXiv

  9. [17]

    Rohan Kshirsagar, Tyus Cukuvac, Kathleen McKeown, and Susan McGregor. 2018. Predictive embeddings for hate speech detection on twitter. arXiv preprint arXiv:1809.10644

    2018 arXiv

  10. [18]

    Varun Kumar, Ashutosh Choudhary, and Eunah Cho. 2020. https://aclanthology.org/2020.lifelongnlp-1.3 Data augmentation using pre-trained transformer models . In Proceedings of the 2nd Workshop on Life-long Learning for Spoken Language Systems, pages 18--26, Suzhou, China. Assoc...

    2020

  11. [19]

    Hongzhan Lin, Ziyang Luo, Wei Gao, Jing Ma, Bo Wang, and Ruichao Yang. 2024. http://arxiv.org/abs/2401.13298 Towards explainable harmful meme detection through multimodal debate between large language models

    2024 arXiv

  12. [20]

    Jiateng Liu, Lin Ai, Zizhou Liu, Payam Karisani, Zheng Hui, Yi Fung, Preslav Nakov, Julia Hirschberg, and Heng Ji. 2025. https://aclanthology.org/2025.coling-main.376/ P ropa I nsight: Toward deeper understanding of propaganda in terms of techniques, appeals, and intent . In P...

    2025

  13. [21]

    Yu Meng, Jiaxin Huang, Yu Zhang, and Jiawei Han. 2022. Generating training data with language models: Towards zero-shot language understanding. Advances in Neural Information Processing Systems, 35:462--477

    2022

  14. [22]

    Xiangyu Qi, Yi Zeng, Tinghao Xie, Pin-Yu Chen, Ruoxi Jia, Prateek Mittal, and Peter Henderson. 2023. Fine-tuning aligned language models compromises safety, even when users do not intend to! arXiv preprint arXiv:2310.03693

    2023 arXiv

  15. [23]

    Gaurav Rajput, Narinder Singh Punn, Sanjay Kumar Sonbhadra, and Sonali Agarwal. 2021. https://doi.org/10.1007/978-3-030-93620-4_6 Hate Speech Detection Using Static BERT Embeddings , page 67–77. Springer International Publishing

    2021 doi

  16. [24]

    Gemma Team, Thomas Mesnard, Cassidy Hardin, Robert Dadashi, Surya Bhupatiraju, Shreya Pathak, Laurent Sifre, Morgane Rivière, Mihir Sanjay Kale, Juliette Love, Pouya Tafti, Léonard Hussenot, Pier Giuseppe Sessa, Aakanksha Chowdhery, Adam Roberts, Aditya Barua, Alex Botev, Alex...

    2024 arXiv

  17. [25]

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. 2023. http://arxiv.org/abs/2302.13971 Llama: ...

    2023 arXiv

  18. [26]

    Jason Wei and Kai Zou. 2019. https://doi.org/10.18653/v1/D19-1670 EDA : Easy data augmentation techniques for boosting performance on text classification tasks . In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International...

    2019 doi

  19. [27]

    BigScience Workshop, :, Teven Le Scao, Angela Fan, Christopher Akiki, Ellie Pavlick, Suzana Ilić, Daniel Hesslow, Roman Castagné, Alexandra Sasha Luccioni, François Yvon, Matthias Gallé, Jonathan Tow, Alexander M. Rush, Stella Biderman, Albert Webson, Pawan Sasanka Ammanamanch...

    2023 arXiv

  20. [28]

    Jiacheng Ye, Jiahui Gao, Jiangtao Feng, Zhiyong Wu, Tao Yu, and Lingpeng Kong. 2022 a . Progen: Progressive zero-shot dataset generation via in-context feedback. arXiv preprint arXiv:2210.12329

    2022 arXiv

  21. [30]

    Jiacheng Ye, Jiahui Gao, Qintong Li, Hang Xu, Jiangtao Feng, Zhiyong Wu, Tao Yu, and Lingpeng Kong. 2022 c . Zerogen: Efficient zero-shot learning via dataset generation. arXiv preprint arXiv:2202.07922

    2022 arXiv

  22. [31]

    Yue Yu, Yuchen Zhuang, Jieyu Zhang, Yu Meng, Alexander J Ratner, Ranjay Krishna, Jiaming Shen, and Chao Zhang. 2024. Large language model as attributed training data generator: A tale of diversity and bias. Advances in Neural Information Processing Systems, 36

    2024

  23. [32]

    Yue Zhang, Hongliang Fei, Dingcheng Li, and Ping Li. 2022. https://doi.org/10.18653/v1/2022.findings-naacl.3 P rompt G en: Automatically generate prompts using generative models . In Findings of the Association for Computational Linguistics: NAACL 2022, pages 30--37, Seattle, ...

    2022 doi

  24. [33]

    Ziqi Zhang, David Robinson, and Jonathan A. Tepper. 2018. https://api.semanticscholar.org/CorpusID:46939253 Detecting hate speech on twitter using a convolution-gru based deep neural network . In Extended Semantic Web Conference

    2018

  25. [34]

    Yaowei Zheng, Richong Zhang, Junhao Zhang, Yanhan Ye, Zheyan Luo, Zhangchi Feng, and Yongqiang Ma. 2024. http://arxiv.org/abs/2403.13372 Llamafactory: Unified efficient fine-tuning of 100+ language models . In Proceedings of the 62nd Annual Meeting of the Association for Compu...

    2024 arXiv

  26. [35]

    Yongchao Zhou, Andrei Ioan Muresanu, Ziwen Han, Keiran Paster, Silviu Pitis, Harris Chan, and Jimmy Ba. 2023. http://arxiv.org/abs/2211.01910 Large language models are human-level prompt engineers

    2023 arXiv

  27. [36]

    URL: " 'urlintro :=

    ENTRY address author booktitle chapter edition editor howpublished institution journal key month note number organization pages publisher school series title type volume year eprint doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before...

  28. [37]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.