Pith. sign in

REVIEW 5 major objections 6 minor 5 cited by

Advancing Content Moderation: Evaluating Large Language Models for Detecting Sensitive Content Across Text, Images, and Videos

T0 review · 5 major / 6 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 The paper claims that general-purpose LLMs—GPT-4o, Gemini 1.5, and Llama-3—detect sensitive content in text, images, and videos with higher accuracy and lower false positive and false negative rates than dedicated moderation APIs and…

desk verdict Useful broad LLM moderation sweep with fresh numbers, but the 'LLMs outperform traditional techniques' headline is not backed by controlled baseline comparisons. read the letter →

arxiv 2411.17123 v1 pith:2NXNPJ4J submitted 2024-11-26 cs.CV cs.AI

Nouar AlDahoul , Myles Joshua Toledo Tan , Harishwar Reddy Kasireddy , Yasir Zaki This is my paper · ORCID
classification cs.CVcs.AI
keywords LargeLanguageModelsVisioncapabilityContentModerationMediaCensorshipViolenceNuditySexualHateSpeech
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 tries to establish that general-purpose multimodal large language models can act as a single content-moderation layer across text, images, and videos. On datasets spanning hate-speech tweets, violence tweets, article descriptions, human photos, cartoons, sketches, graphic-violence images, and surveillance and real-life violence videos, the authors report that prompt-driven GPT-4o, Gemini 1.5, and Llama-3 models achieve higher accuracy and lower false positive and false negative rates than dedicated moderation APIs and older CNN/LSTM detectors. The paper also documents where dedicated tools fail, including graphic violence, alcohol/drug/child abuse, and long news articles. If the results hold, platforms could replace several specialized classifiers with one multimodal model and reduce both over-blocking and missed harmful content.

What carries the argument

The operative mechanism is closed-set classification by prompted generation: each model is given a fixed category list, such as ['violence','fighting','normal'] or ['nude','porn','naked','sex','normal'], and asked to answer with a single word. For images and video frames, the task is posed as visual question answering; videos are converted to frame sequences by taking one frame out of every ten and feeding the frames as multiple inputs. Because models have native safety taxonomies, the evaluation bridges their outputs to the prompt categories with explicit equivalences: Gemini's 'dangerous' category counts as violence and harm, low-probability Gemini blocks count as normal, and Llama-3's 'harassment' category counts as adult content in the article-description task. These equivalences are load-bearing parts of the reported numbers.

What would settle it

Re-run the same textual and visual benchmarks using raw model outputs without any manual relabeling, and compare against the paper's numbers; the adult-content task already shows the sensitivity, since Llama-3.1-8B-Instruct's accuracy drops from 94.24% to 81.41% when 'harassment' is not counted as adult.

Watch

Extended reading notes

Core claim

The central claim, stated in the abstract and conclusion, is that LLMs outperform traditional techniques by achieving higher accuracy and lower false positive and false negative rates. The evidence comes from closed-set classification prompts applied to text, single images, and sampled video frames: for violence in videos, Gemini 1.5 Pro reaches 95.5% accuracy and balances FPR and FNR at 4.5%; for nudity in human photos, Llama-3.2-11B-Vision-Instruct reaches 97.63% accuracy; for cartoon nudity, Gemini 1.5 Pro reaches 99.89%; and for graphic violence, Llama-3.2-11B-Vision-Instruct reaches 98.57%. No single model dominates every task: dedicated tools are more competitive for hate speech and offensive language, while general LLMs fill gaps in graphic violence and abuse detection that the dedicated moderators cannot cover at all.

Load-bearing premise

The reported accuracies depend on the authors' manual equivalences between each model's native safety labels and the prompt categories, such as counting Gemini's 'dangerous' as violence and Llama-3's 'harassment' as adult content.

Editorial extensions

If this is right

  • Platforms could use one multimodal LLM or a small set of them to moderate text, image, and video with a single prompt pipeline, replacing separate classifiers for hate speech, nudity, and violence.
  • Because the best model varies by task, a production system would route content by type: Gemini-class models for visual censorship, GPT-4o-class models for violent text, and the OpenAI moderation model for hate speech and offensive language.
  • The dedicated moderation APIs' failure on graphic violence and abuse categories means safety-tuned general LLMs cover a wider range of harms without task-specific retraining.
  • LLM predictions on unlabeled news and review corpora can serve as a monitor for how much inappropriate content is present and how it trends over time, as demonstrated by the 2013-2023 news-article analysis.

Reading between the lines

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

  • The paper's performance comparisons inherit its category-equivalence choices; a deployment that used literal model outputs would likely see smaller advantages for Gemini and Llama-3 on violence and adult-content tasks.
  • A practical ensemble could exploit the observed model disagreement: flag content when any model in the set flags it, so the higher false-positive load is sent to human review rather than letting harmful content through.
  • A testable extension is to replace the fixed category prompts with free-text justifications and measure whether explanation quality tracks detection accuracy, separating model judgment from output-taxonomy translation.
  • The news-trend analysis suggests the same prompts can be rerun over time as a low-cost content-prevalence dashboard without needing new labels.
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

5 major / 6 minor

Summary. This paper evaluates a set of LLMs (GPT-4o, GPT-4o-mini, Gemini 1.5 Pro/Flash, Llama-3.1-8B-Instruct, Llama-Guard-3, and OpenAI moderation model) on content-moderation tasks across text, images, and videos. For text, it uses hate-speech/offensive-language tweets, gender-based-violence tweets, adult-content article descriptions, Amazon reviews, and news articles; for visual content, it uses human-photo, cartoon, sketch, violence-video, graphic-violence, and abuse datasets. The main claims are that LLMs outperform traditional content-moderation techniques and specialized moderators in accuracy and in balancing false positives and false negatives, and that general-purpose multimodal LLMs can serve as effective censors across modalities.

Significance. If the headline result were established, the paper would provide a useful practical benchmark for content moderation and would support the use of general-purpose multimodal LLMs as moderation backends. The breadth of the evaluation, spanning five text sources, four image collections, and an 800-video test set, is a strength, as is the explicit reporting of prompts and the treatment of video via multiple frames. However, the comparative claim against traditional techniques is not supported by the experimental design: no baselines are run for text, and the visual baselines are imported from other papers without matching splits or metrics. The label-equivalence decisions and the absence of uncertainty quantification further limit the reliability of the reported numbers. The paper is therefore better read as a broad LLM-to-LLM comparison and a set of use-case studies than as a demonstration that LLMs outperform traditional methods.

major comments (5)
  1. [Abstract / Section 6 / Section 5.2] The abstract and the conclusion (Section 6) state that 'LLMs outperform traditional techniques by achieving higher accuracy and lower false positive and false negative rates,' but the experiments do not evaluate any traditional baseline under the same protocol. For the textual tasks (Sections 5.2.1-5.2.3), no SVM, TF-IDF, BERT, or CNN/LSTM baseline is run; the comparisons are only among LLMs. For the visual tasks, Tables 1-4 take baseline numbers from the cited papers, and Table 1 explicitly says these numbers are 'mentioned in their main articles' while the LLMs are evaluated on an 800-video subset of the 4,000-video RWF-2000/RLVS pool; moreover, several FNR cells are marked 'X.' As a result, the reported accuracy/FPR/FNR gaps are not controlled deltas on identical data and metrics, and the headline claim is unsupported.
  2. [Section 5 / Section 5.2.1] Several reported accuracies depend on author-defined equivalences that are introduced during result analysis. Section 5.1.1 treats Gemini low-probability blocks as 'normal' content, Section 5.1.2 maps Gemini's 'dangerous' category onto violence, Section 5.2 maps Gemini low-probability text blocks to 'normal,' and Section 5.2.1 counts Llama-3's 'harassment' outputs as adult content. The paper itself reports that without this last mapping, Llama-3's accuracy on adult-content description drops from 94.24% to 81.41% and recall from 94.67% to 53.55%. These mappings are post hoc and should be justified, ablated, or treated as sensitivity analyses before the numbers are presented as model accuracy.
  3. [Section 4.1.2 / Table 5] The graphic violence and abuse datasets are created by the authors by scraping images from the web, but the paper provides no annotation protocol, no label verification, and no indication of how the ground-truth labels were assigned. Without a documented and validated labeling process, the per-category accuracy values in Table 5 (and the claim in Section 5.1.3 that Llama-Guard-3 'is not able to detect graphic violence') are not reliable.
  4. [All tables / Section 5] The results are reported as point estimates from a single pass; the paper does not mention repeated runs, temperature settings, seeds, or confidence intervals for any of the stochastic LLM baselines. Given that the models are sampled (and that Section 5.1.2 describes changing prompts for unanswered samples), the differences between models (e.g., 95.5% vs. 94.13% in Table 1) could be within run-to-run variance, and no statistical significance can be assessed.
  5. [Sections 5.2.2-5.2.5] For the HSOL, GBV, Amazon-review, and news-article tasks, the results are presented only as heatmaps of predicted categories or per-category percentages; no aggregate accuracy, precision, recall, F1, FPR, or FNR values are reported against ground truth (which exists for HSOL and GBV). Consequently, statements such as 'OpenAI moderation model shows the best capability for the task of hate and offensive language detection' (Section 5.2.2) are not backed by quantitative metrics.
minor comments (6)
  1. [Table 4] In the GPT-4o-mini row, the FNR is listed as '12.2%%'; the extra percent sign should be removed.
  2. [Section 5.1.1] The paper says an 800-video test subset was used from the combined RWF-2000/RLVS pool, but it does not describe how the subset was sampled or whether the cited baselines were evaluated on the same subset; this should be clarified.
  3. [Section 5.2] The prompt for GPT-4o and Gemini 1.5 asks for a confidence score between 0 and 10, but the paper never explains how these scores are used (thresholds, aggregation, or exclusion) in computing the reported metrics.
  4. [Section 5.1.2] The prompt change for Gemini 1.5 Pro on unanswered nudity images ('Does this image have NSFW content?') is mentioned only in prose; the number of affected samples and the resulting label distribution should be reported for reproducibility.
  5. [Section 4] The prompts are embedded in the text with no consistent numbering; a summary table or appendix listing all prompts would improve reproducibility.
  6. [References] Tables 2-4 list AWS and Microsoft Azure numbers from reference [15], but it is not stated whether these cloud services were re-run under the same image set or transcribed from that paper; if transcribed, the original protocol should be summarized.

Circularity Check

3 steps flagged · score 4.0 of 10

Label-equivalence choices re-define some LLM outputs as correct (Llama-3 'harassment' as adult; Gemini block levels as labels), but the central comparison retains independent content.

  1. self definitional [Section 5.2.1, Detection of Adult Content in Articles' Description (text before Table 6)]
    "For Llama-3, in the adult content detection of the articles' descriptions, the 'harassment' category was generated for several samples; even the 'sexual' category is available as an option in the Llama-3's prompt. If this 'harassment' category is considered an 'adult' category, the performance metrics of Llama-3 are shown in Table 6. If we consider the 'harassment' category as a 'non-adult' category because the model fails to select the 'sexual' category from the options given, the accuracy would degrade largely to 81.41% with a low recall of 53.55% and a high FNR of 46.44%."

    The reported 94.24% accuracy for Llama-3 is obtained only after defining the ground-truth 'adult' class to include the model's own 'harassment' output. The dataset provides adult/non-adult labels; the equivalence 'harassment = adult' is not given by the dataset but is chosen post hoc because Llama-3 produced 'harassment' instead of 'sexual'. The paper itself shows the accuracy drops to 81.41% under the alternative coding, demonstrating that the headline number is an artifact of the re-labeling. The prediction is scored against a target definition that was adjusted to fit the model's confusion.

  2. self definitional [Section 5.2, Textual Content Analysis (Gemini safety-setting mapping before Section 5.2.1)]
    "For Gemini 1.5, we set the safety settings to be 'BLOCK-LOW-AND-ABOVE'. In this case, Gemini 1.5 will block prompts when the probability score is either low, medium, or high. In this experiment, the text samples that were blocked with medium and high probability scores are considered to have adult content. On the other hand, the samples that were blocked with low probability scores and the unblocked samples are predicted under the 'normal' category."

    Here the predicted class for Gemini is read directly from the model's safety-block severity: medium/high blocks are coded as adult, while low-probability blocks and unblocked samples are coded as normal. No independent validation of this equivalence is provided. Because the model's own internal threshold determines which ground-truth labels are assigned, Gemini's reported accuracy in the textual adult-content task is constructed by the evaluation mapping rather than by an independent test of the model's outputs against the dataset labels.

1 more flagged steps
  1. self definitional [Section 5.1.1, Detection of Violence in Videos (Gemini block-level mapping for video violence)]
    "The videos blocked with low probability scores are considered to have normal content, while others blocked with medium or high probability scores are categorized as having violent content."

    The same construction is applied to the video violence task: the ground-truth label for Gemini is defined by the model's block level (low-probability block = normal; medium/high block = violence). This means Gemini 1.5 Pro's 95.5% accuracy in Table 1 is partly an artifact of mapping the model's safety refusal levels onto the dataset labels rather than a direct comparison with the video labels. Since the block level is produced by the model being evaluated, the reported performance is partially self-referential.

full rationale

The paper's central claim ('LLMs outperform traditional techniques') is not structurally circular in its entirety: most LLM numbers come from prompting models on external datasets and comparing with dataset labels, and the visual baselines are imported from prior published work (some by the same authors) rather than re-fitted in this paper. Those imported baselines are a fairness/control limitation, not a circularity, because the baseline numbers are external and not derived from the present paper's assumptions. Similarly, self-citations such as [72], [75], [77], [87], and [88] supply independent published results that were not produced by the current evaluation; their use does not make the claim circular. However, two evaluation mappings are circular in the narrow sense. First, Llama-3's 'harassment' outputs are re-labelled as correct 'adult' detections, and the paper itself shows the accuracy depends on this coding (94.24% vs 81.41%). Second, Gemini's own safety-block severity (low/medium/high) is used to define whether a sample is 'normal' or contains adult/violent content, so the model's internal refusal threshold is converted into ground-truth labels. These steps affect specific reported accuracies (e.g., Llama-3 in Table 6, Gemini in Tables 1 and related textual results) and therefore warrant a moderate score, but they do not force the entire conclusion, since direct prompt-based evaluations of GPT-4o, GPT-4o-mini, and Llama-3.2-11B-Vision also support strong LLM performance in several tasks. Overall score 4.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new entities. Its central claim rests on benchmark data, prompt design choices, and label-mapping decisions, which are listed above as free parameters and axioms.

free parameters (4)
  • Gemini low-probability block mapping = low probability block (normal), medium/high block (positive class)
    Hand-set decision rule in Section 5.1 that determines accuracy for Gemini on violence and text tasks; no sensitivity analysis is provided.
  • Llama-3 harassment-to-adult relabeling = harassment treated as adult category in Table 6
    Post hoc mapping applied only to the adult-content task; authors note accuracy drops from 94.24% to 81.41% otherwise (Section 5.2.1).
  • Video frame sampling rate = one frame per ten frames
    Sampling choice in Section 5.1.1 that could affect violence detection; no ablation is reported.
  • Gemini safety settings = BLOCK-LOW-AND-ABOVE for violence; BLOCK-NONE for nudity
    Per-task settings chosen by the authors, and the prompt was changed mid-experiment when Gemini 1.5 Pro refused to answer some images (Section 5.1).
assumptions (5)
  • domain assumption Custom scraped datasets (350 graphic violence images, 423 abuse images) are accurately labeled as positive examples.
    Section 4.1.2 describes scraping without expert annotation or inter-annotator agreement; accuracy results in Table 5 depend on label correctness.
  • domain assumption Baseline metrics from prior papers are comparable despite differing test splits and evaluation protocols.
    Section 5.1.1 and Tables 1-4 quote baseline numbers from earlier papers without re-running them on the same test sets.
  • ad hoc to paper Model output categories are semantically alignable to the prompt label sets.
    The authors map Gemini 'dangerous' to violence and Llama-3 'harassment' to adult content in Section 5; these equivalences are not validated.
  • domain assumption For the unlabeled news and Amazon review corpora, model predictions reflect actual content prevalence.
    Sections 5.2.4 and 5.2.5 answer RQ4 using model output distributions only; there is no ground truth for these corpora.
  • domain assumption LLM responses are treated as deterministic; single API calls per sample.
    No repeated sampling or temperature settings are reported, though API outputs are stochastic in general.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Advancing Content Moderation: Evaluating Large Language Models for Detecting Sensitive Content Across Text, Images, and Videos." pith.science (2026). https://pith.science/paper/2NXNPJ4J

@misc{pith2026241117123,
  author       = {Pith},
  title        = {Pith review of: Advancing Content Moderation: Evaluating Large Language Models for Detecting Sensitive Content Across Text, Images, and Videos},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2NXNPJ4J}},
  note         = {Machine review of arXiv:2411.17123}
}
read the original abstract

The widespread dissemination of hate speech, harassment, harmful and sexual content, and violence across websites and media platforms presents substantial challenges and provokes widespread concern among different sectors of society. Governments, educators, and parents are often at odds with media platforms about how to regulate, control, and limit the spread of such content. Technologies for detecting and censoring the media contents are a key solution to addressing these challenges. Techniques from natural language processing and computer vision have been used widely to automatically identify and filter out sensitive content such as offensive languages, violence, nudity, and addiction in both text, images, and videos, enabling platforms to enforce content policies at scale. However, existing methods still have limitations in achieving high detection accuracy with fewer false positives and false negatives. Therefore, more sophisticated algorithms for understanding the context of both text and image may open rooms for improvement in content censorship to build a more efficient censorship system. In this paper, we evaluate existing LLM-based content moderation solutions such as OpenAI moderation model and Llama-Guard3 and study their capabilities to detect sensitive contents. Additionally, we explore recent LLMs such as GPT, Gemini, and Llama in identifying inappropriate contents across media outlets. Various textual and visual datasets like X tweets, Amazon reviews, news articles, human photos, cartoons, sketches, and violence videos have been utilized for evaluation and comparison. The results demonstrate that LLMs outperform traditional techniques by achieving higher accuracy and lower false positive and false negative rates. This highlights the potential to integrate LLMs into websites, social media platforms, and video-sharing services for regulatory and content moderation purposes.

Figures

Figures reproduced from arXiv: 2411.17123 by the authors.

Figure 1
Figure 1. The Block Diagram of our Proposed Solution [PITH_FULL_IMAGE:figures/full_fig_p015_1.png] view at source ↗
Figure 2
Figure 2. Heatmap of normal text showing predictions of various LLMs. [PITH_FULL_IMAGE:figures/full_fig_p033_2.png] view at source ↗
Figure 3
Figure 3. Heatmap of hate speech showing predictions of various LLMs. [PITH_FULL_IMAGE:figures/full_fig_p033_3.png] view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: Heatmap of offensive language showing predictions of various LLMs. [PITH_FULL_IMAGE:figures/full_fig_p033_4.png]
Figure 5
Figure 5. Figure 5: Heatmap of sexual violence showing predictions of various LLMs. [PITH_FULL_IMAGE:figures/full_fig_p035_5.png]
Figure 6
Figure 6. Figure 6: Heatmap of physical violence showing predictions of various LLMs. [PITH_FULL_IMAGE:figures/full_fig_p035_6.png]
Figure 7
Figure 7. Figure 7: Heatmap of emotional violence showing predictions of various LLMs. [PITH_FULL_IMAGE:figures/full_fig_p036_7.png]
Figure 8
Figure 8. Figure 8: Heatmap of economic violence showing predictions of various LLMs. [PITH_FULL_IMAGE:figures/full_fig_p036_8.png]
Figure 9
Figure 9. Figure 9: Heatmap of harmful practice showing predictions of various LLMs. [PITH_FULL_IMAGE:figures/full_fig_p036_9.png]
Figure 10
Figure 10. Figure 10: Heatmap of inappropriate content distribution in the Amazon beauty reviews. [PITH_FULL_IMAGE:figures/full_fig_p037_10.png]
Figure 11
Figure 11. Figure 11: Heatmap of inappropriate content distribution in the CNN articles. [PITH_FULL_IMAGE:figures/full_fig_p038_11.png]
Figure 12
Figure 12. Figure 12: Heatmap of inappropriate content distribution in the Fox News articles. [PITH_FULL_IMAGE:figures/full_fig_p039_12.png]
Figure 13
Figure 13. Figure 13: Heatmap of inappropriate content distribution in the Newsweek articles. [PITH_FULL_IMAGE:figures/full_fig_p039_13.png]
Figure 14
Figure 14. Figure 14: Heatmap of inappropriate content distribution in the Daily Beast articles. [PITH_FULL_IMAGE:figures/full_fig_p040_14.png]
Figure 15
Figure 15. Figure 15: Heatmap of inappropriate content distribution in the Washington Times arti [PITH_FULL_IMAGE:figures/full_fig_p041_15.png]
Figure 16
Figure 16. Figure 16: Trends of inappropriate content in media outlets: A comparative analysis of [PITH_FULL_IMAGE:figures/full_fig_p042_16.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 5 Pith papers

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

  1. Beyond Binary Moderation: Identifying Fine-Grained Sexist and Misogynistic Behavior on GitHub with Large Language Models

    cs.SE 2025-07 conditional novelty 6.0 of 10

    An instruction-tuned GPT-4o prompt achieves an MCC of 0.501 on 12-category sexism/misogyny classification of GitHub comments, but the evaluation was tuned on the same test set.

  2. Behind Closed Words: Creating and Investigating the forePLay Annotated Dataset for Polish Erotic Discourse

    cs.CL 2024-12 conditional novelty 6.0 of 10

    A new 24,768-sentence Polish dataset labels erotic and related content, and Polish-specific language models outperform multilingual models at detecting it.

  3. VModA: An Effective Framework for Adaptive NSFW Image Moderation

    cs.CV 2025-05 conditional novelty 5.0 of 10

    VModA combines prompt engineering, region zooming, and LLM-based answer aggregation to improve zero-shot NSFW image moderation across multiple categories.

  4. Dynamic Content Moderation in Livestreams: Combining Supervised Classification with MLLM-Boosted Similarity Matching

    cs.CV 2025-12 conditional novelty 4.0 of 10

    A deployed hybrid moderation system combining supervised classification and reference-based similarity matching, boosted by MLLM distillation, reduces unwanted livestream views by 6–8%.

  5. Benchmarking the Legal Reasoning of LLMs in Arabic Islamic Inheritance Cases

    cs.CL 2025-08 conditional novelty 3.0 of 10

    A majority-vote ensemble of Gemini Flash 2.5, Gemini Pro 2.5, and GPT o3 achieves 92.7% accuracy on the QIAS 2025 Islamic inheritance test set, outperforming each model alone and all open Arabic models.

Reference graph

Works this paper leans on

118 extracted references · 68 canonical work pages · cited by 5 Pith papers

  1. [1]

    Justice and revenge in online counter-publics: Emerging responses to sexual violence in the age of social media.Crime, Media, Culture 9, 225–242 (2013)

    Salter, M. Justice and revenge in online counter-publics: Emerging responses to sexual violence in the age of social media.Crime, Media, Culture 9, 225–242 (2013)

    2013

  2. [2]

    C., Jordan, A

    Strasburger, V. C., Jordan, A. B. & Donnerstein, E. Health effects of media on children and adolescents.Pediatrics 125, 756–767 (2010)

    2010

  3. [3]

    Greenfield, P. M. Inadvertent exposure to pornography on the internet: Implications of peer-to-peer file-sharing networks for child development and families. Journal of Applied Developmental Psychology25, 741– 750 (2004)

    2004

  4. [4]

    Citron, D. K. Sexual privacy.Yale LJ 128, 1870 (2018)

    2018

  5. [5]

    Desensitization to media violence: links with habitual media violence exposure, aggressive cognitions, and aggressive behav- ior

    Krahé, B.et al. Desensitization to media violence: links with habitual media violence exposure, aggressive cognitions, and aggressive behav- ior. Journal of personality and social psychology100, 630 (2011)

    2011

  6. [6]

    https://www.justice.gov/criminal/ criminal-ceos/child-pornography

    Child pornography. https://www.justice.gov/criminal/ criminal-ceos/child-pornography

  7. [7]

    & Henry, N

    Powell, A. & Henry, N. Sexual violence in a digital age (Springer, 2017)

    2017

  8. [8]

    Erin Simpson, A. C. How to regulate tech: A technology policy framework for online ser- vices. https://www.americanprogress.org/article/ how-to-regulate-tech-a-technology-policy-framework-for-online-services

Show all 118 references
  1. [9]

    L., Giovanni Sartor

    Giovanni Sartor, A. L., Giovanni Sartor. The impact of al- gorithms for online content filtering or moderation. https: //www.europarl.europa.eu/RegData/etudes/STUD/2020/657101/ IPOL_STU(2020)657101_EN.pdf

    2020

  2. [10]

    https:// newmajorityfoundation.com/freedom-of-speech/

    Freedom of speech in the digital age. https:// newmajorityfoundation.com/freedom-of-speech/

  3. [11]

    Moderation

    platform, O. Moderation. https://platform.openai.com/docs/ guides/moderation. 43

  4. [12]

    A holistic approach to undesired content detection in the real world

    Markov, T.et al. A holistic approach to undesired content detection in the real world. InProceedings of the AAAI Conference on Artificial Intelligence, vol. 37, 15009–15018 (2023)

    2023

  5. [13]

    Inan, H. et al. Llama guard: Llm-based input-output safeguard for human-ai conversations.arXiv preprint arXiv:2312.06674(2023)

    2023 arXiv

  6. [14]

    Llama Team, A. . M. The llama 3 herd of models (2024). URLhttps: //arxiv.org/abs/2407.21783. 2407.21783

    2024 arXiv

  7. [15]

    A., Momo, M

    AlDahoul, N., Karim, H. A., Momo, M. A., Sy, M. A. & Tan, M. J. T. Evaluation of content moderation software for nudity and pornography detection in various scenarios.MECON Multimedia University Engi- neering Conference(2023)

    2023

  8. [16]

    Jahan, M. S. & Oussalah, M. A systematic review of hate speech au- tomatic detection using natural language processing.Neurocomputing 546, 126232 (2023)

    2023

  9. [17]

    & Picard, R

    Dinakar, K., Jones, B., Havasi, C., Lieberman, H. & Picard, R. Com- mon sense reasoning for detection, prevention, and mitigation of cyber- bullying. ACM Transactions on Interactive Intelligent Systems (TiiS) 2, 1–30 (2012)

    2012

  10. [18]

    Abozinadah, E. A. & Jones, J. H. Improved micro-blog classification for detecting abusive arabic twitter accounts. International Journal of Data Mining & Knowledge Management Process (IJDKP)6, 17–28 (2016)

    2016

  11. [19]

    & Varma, V

    Badjatiya, P., Gupta, S., Gupta, M. & Varma, V. Deep learning for hate speech detection in tweets. InProceedings of the 26th international conference on World Wide Web companion, 759–760 (2017)

    2017

  12. [20]

    & Weber, I

    Davidson, T., Warmsley, D., Macy, M. & Weber, I. Automated hate speech detection and the problem of offensive language. InProceedings of the international AAAI conference on web and social media, vol. 11, 512–515 (2017)

    2017

  13. [21]

    & Raje, R

    Pawar, R., Agrawal, Y., Joshi, A., Gorrepati, R. & Raje, R. R. Cy- berbullying detection system with multiple server configurations. In 44 2018 IEEE International Conference on Electro/Information Technol- ogy (EIT), 0090–0095 (IEEE, 2018)

    2018

  14. [22]

    & Yeung, D.-Y

    Ousidhoum, N., Lin, Z., Zhang, H., Song, Y. & Yeung, D.-Y. Multilingual and multi-aspect hate speech analysis. arXiv preprint arXiv:1908.11049 (2019)

    2019 arXiv

  15. [23]

    & Nikolov, N

    Alakrot, A., Murray, L. & Nikolov, N. S. Towards accurate detec- tion of offensive language in online communication in arabic.Procedia computer science 142, 315–320 (2018)

    2018

  16. [24]

    & Zampieri, M

    Malmasi, S. & Zampieri, M. Challenges in discriminating profanity from hate speech. Journal of Experimental & Theoretical Artificial Intelligence 30, 187–202 (2018)

    2018

  17. [25]

    & Joshi, A

    Kamble, S. & Joshi, A. Hate speech detection from code-mixed hindi-english tweets using deep learning models. arXiv preprint arXiv:1811.05145 (2018)

    2018 arXiv

  18. [26]

    & Castillo, P

    Faris, H., Aljarah, I., Habib, M. & Castillo, P. A. Hate speech detec- tion using word embedding and deep learning in the arabic language context. In ICPRAM, 453–460 (2020)

    2020

  19. [27]

    & Schuller, B

    Rizos, G., Hemker, K. & Schuller, B. Augment to prevent: short-text data augmentation in deep learning for hate-speech classification. In Proceedings of the 28th ACM international conference on information and knowledge management, 991–1000 (2019)

    2019

  20. [28]

    & Savage, N

    Zhou, Y., Yang, Y., Liu, H., Liu, X. & Savage, N. Deep learning based fusion approach for hate speech detection. IEEE Access 8, 128923– 128929 (2020)

    2020

  21. [29]

    & Mamidi, R

    Dowlagar, S. & Mamidi, R. Hasocone@ fire-hasoc2020: Using bert and multilingual bert models for hate speech detection.arXiv preprint arXiv:2101.09007 (2021)

    2021 arXiv

  22. [30]

    Mulki, H., Haddad, H., Ali, C. B. & Alshabani, H. L-hsab: A levantine twitter dataset for hate speech and abusive language. InProceedings of the third workshop on abusive language online, 111–118 (2019). 45

    2019

  23. [31]

    & Schütze, H

    Yin, W., Kann, K., Yu, M. & Schütze, H. Comparative study of cnn and rnn for natural language processing. arXiv preprint arXiv:1702.01923 (2017)

    2017 arXiv

  24. [32]

    K., Ramampiaro, H

    Pitsilis, G. K., Ramampiaro, H. & Langseth, H. Effective hate-speech detection in twitter data using recurrent neural networks.Applied In- telligence 48, 4730–4742 (2018)

    2018

  25. [33]

    S., Alhothali, A

    Alatawi, H. S., Alhothali, A. M. & Moria, K. M. Detecting white supremacist hate speech using domain specific word embedding with deep learning and bert.IEEE Access 9, 106363–106374 (2021)

    2021

  26. [34]

    & Hettiarachchi, H

    Ranasinghe, T., Zampieri, M. & Hettiarachchi, H. Brums at hasoc 2019: Deep learning models for multilingual hate speech and offensive language identification. InFIRE (working notes), 199–207 (2019)

    2019

  27. [35]

    & Semeraro, G

    Polignano, M., Basile, V., Basile, P., de Gemmis, M. & Semeraro, G. Alberto: Modeling italian social media language with bert. IJCoL. Italian Journal of Computational Linguistics5, 11–31 (2019)

    2019

  28. [36]

    InProceedings of the First Work- shop on Bangla Language Processing (BLP-2023), 236–240 (2023)

    Das, R.et al.Team errorpoint atblp-2023 task1: Acomprehensiveap- proach for violence inciting text detection using deep learning and tra- ditional machine learning algorithm. InProceedings of the First Work- shop on Bangla Language Processing (BLP-2023), 236–240 (2023)

    2023

  29. [37]

    & Sonawane, S

    Page, S., Mangalvedhekar, S., Deshpande, K., Chavan, T. & Sonawane, S. Mavericks at blp-2023 task 1: Ensemble-based approach using lan- guage models for violence inciting text detection. In Proceedings of the first workshop on bangla language processing (BLP-2023), 190–195 (2023)

    2023

  30. [38]

    S., Malik, M

    Khan, M. S., Malik, M. S. I. & Nadeem, A. Detection of violence inci- tation expressions in urdu tweets using convolutional neural network. Expert Systems with Applications245, 123174 (2024)

    2024

  31. [39]

    Ba Wazir, A. S.et al. Design and implementation of fast spoken foul language recognition with different end-to-end deep neural network ar- chitectures. Sensors 21, 710 (2021). 46

    2021

  32. [40]

    Wazir, A. S. B.et al. Spectrogram-based classification of spoken foul language using deep cnn. In2020 IEEE 22nd International Workshop on Multimedia Signal Processing (MMSP), 1–6 (IEEE, 2020)

    2020

  33. [41]

    T., Wilson, C

    Nguyen, T. T., Wilson, C. & Dalins, J. Fine-tuning llama 2 large lan- guage models for detecting online sexual predatory chats and abusive texts. arXiv preprint arXiv:2308.14683(2023)

    2023 arXiv

  34. [42]

    & Akbik, A

    Vogt, M., Leser, U. & Akbik, A. Early detection of sexual predators in chats. InProceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), 4985–4999 (2021)

    2021

  35. [43]

    Hamzah, N. A. & Dhannoon, B. N. The detection of sexual harassment and chat predators using artificial neural network.Karbala Interna- tional Journal of Modern Science7, 6 (2021)

    2021

  36. [44]

    & Luo, X

    Yan, M. & Luo, X. Bert-based detection of sexual harassment in di- alogues. In Proceedings of the 2021 5th International Conference on Computer Science and Artificial Intelligence, 359–364 (2021)

    2021

  37. [45]

    & Chen, X

    Ketsbaia, L., Issac, B. & Chen, X. Detection of hate tweets using machine learning and deep learning. In2020 IEEE 19th International Conference on Trust, Security and Privacy in Computing and Commu- nications (TrustCom), 751–758 (IEEE, 2020)

    2020

  38. [46]

    & Al Aghbari, Z

    Khezzar, R., Moursi, A. & Al Aghbari, Z. arhatedetector: detection of hate speech from standard and dialectal arabic tweets. Discover Internet of Things 3, 1 (2023)

    2023

  39. [47]

    & Al-Dossari, H

    Al-Hassan, A. & Al-Dossari, H. Detection of hate speech in arabic tweets using deep learning.Multimedia systems 28, 1963–1974 (2022)

    2022

  40. [48]

    Activelearningto measure opinion and violence in french newspapers.Procedia Computer Science 192, 202–211 (2021)

    Guélorget, P., Gadek, G., Zaharia, T.&Grilheres, B. Activelearningto measure opinion and violence in french newspapers.Procedia Computer Science 192, 202–211 (2021)

    2021

  41. [49]

    J., Palomar, N., Gallego, E., Navascués, L

    Bello, H. J., Palomar, N., Gallego, E., Navascués, L. J. & Lozano, C. Machine learning to study the impact of gender-based violence in the news media. arXiv preprint arXiv:2012.07490(2020). 47

    2020 arXiv

  42. [50]

    & Tesconi, M

    Del Vigna12, F., Cimino23, A., Dell’Orletta, F., Petrocchi, M. & Tesconi, M. Hate me, hate me not: Hate speech detection on face- book. In Proceedings of the first Italian conference on cybersecurity (ITASEC17), 86–95 (2017)

    2017

  43. [51]

    L.et al.Identifying hate speech and attribution of responsibility: An analysis of simulated whatsapp conversations dur- ing the pandemic

    Palazón-Fernández, J. L.et al.Identifying hate speech and attribution of responsibility: An analysis of simulated whatsapp conversations dur- ing the pandemic. InHealthcare, vol. 11, 1564 (MDPI, 2023)

    2023

  44. [52]

    Ries, C. X. & Lienhart, R. A survey on visual adult image recognition. Multimedia tools and applications69, 661–688 (2014)

    2014

  45. [53]

    & da Silva Eleuterio, P

    de Castro Polastro, M. & da Silva Eleuterio, P. M. A statistical ap- proach for identifying videos of child pornography at crime scenes. In 2012 Seventh International Conference on Availability, Reliability and Security, 604–612 (IEEE, 2012)

    2012

  46. [54]

    Arentz, W. A. & Olstad, B. Classifying offensive sites based on im- age content. Computer Vision and Image Understanding94, 295–310 (2004)

    2004

  47. [55]

    A., Ahmad, N

    Zaidan, A., Karim, H. A., Ahmad, N. N., Zaidan, B. B. & Kiah, M. M. Robust pornography classification solving the image size varia- tion problem based on multi-agent learning.Journal of Circuits, Sys- tems and Computers 24, 1550023 (2015)

    2015

  48. [56]

    Z., Wiederhold, G

    Wang, J. Z., Wiederhold, G. & Firschein, O. System for screening objectionable images using daubechies’ wavelets and color histograms. In Interactive Distributed Multimedia Systems and Telecommunication Services: 4th International Workshop, IDMS’97 Darmstadt, Germany, Septembe...

    1997

  49. [57]

    & Gao, W

    Zheng, Q.-F., Zeng, W., Wang, W.-Q. & Gao, W. Shape-based adult image detection.International Journal of Image and Graphics6, 115– 124 (2006)

    2006

  50. [58]

    M., Forsyth, D

    Fleck, M. M., Forsyth, D. A. & Bregler, C. Finding naked people. In Computer Vision—ECCV’96: 4th European Conference on Computer Vision Cambridge, UK, April 15–18, 1996 Proceedings Volume II 4, 593–602 (Springer, 1996). 48

    1996

  51. [59]

    C., Chan, Y

    Bosson, A., Cawley, G. C., Chan, Y. & Harvey, R. Non-retrieval: blocking pornographic images. InInternational Conference on Image and Video Retrieval, 50–60 (Springer, 2002)

    2002

  52. [60]

    J., Kim, J.-s., Nam, T

    Kim, W., Yoo, S. J., Kim, J.-s., Nam, T. Y. & Yoon, K. Detecting adult images using seven mpeg-7 visual descriptors. InWeb and Communi- cation Technologies and Internet-Related Social Issues-HSI 2005: 3rd International Conference on Human. Society@ Internet, Tokyo, Japan, July...

    2005

  53. [61]

    Wijaya, I. G. P. S., Widiartha, I., Uchimura, K. & Koutaki, G. Phono- graphic image recognition using fusion of scale invariant descriptor. In 2015 21st Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV), 1–5 (IEEE, 2015)

    2015

  54. [62]

    & Hauke, R

    Lienhart, R. & Hauke, R. Filtering adult image content with topic models. In 2009 IEEE International Conference on Multimedia and Expo, 1472–1475 (IEEE, 2009)

    2009

  55. [63]

    & Araújo, A

    Caetano, C., Avila, S., Guimaraes, S. & Araújo, A. d. A. Pornography detection using bossanova video descriptor. In 2014 22nd European Signal Processing Conference (EUSIPCO), 1681–1685 (IEEE, 2014)

    2014

  56. [64]

    R., Guimarães, S

    Caetano, C., Avila, S., Schwartz, W. R., Guimarães, S. J. F. & Araújo, A. d. A. A mid-level video representation based on binary descriptors: A case study for pornography detection.Neurocomputing 213, 102–114 (2016)

    2016

  57. [65]

    & Tan, X

    Jin, X., Wang, Y. & Tan, X. Pornographic image recognition via weighted multiple instance learning.IEEE transactions on cybernetics 49, 4412–4420 (2018)

    2018

  58. [66]

    Pornographic image recognition based on multi- instance deep learning.SSRN (2022)

    Li, D., Ji, Z.et al. Pornographic image recognition based on multi- instance deep learning.SSRN (2022)

    2022

  59. [67]

    Perez, M. et al. Video pornography detection through deep learning techniques and motion information. Neurocomputing 230, 279–293 (2017)

    2017

  60. [68]

    Applying deep learning to classify pornographic images and videos

    Moustafa, M. Applying deep learning to classify pornographic images and videos. arXiv preprint arXiv:1511.08899(2015). 49

    2015 arXiv

  61. [69]

    Hor, S. L. et al. An evaluation of state-of-the-art object detectors for pornography detection. In2021 IEEE International Conference on Signal and Image Processing Applications (ICSIPA), 191–196 (IEEE, 2021)

    2021

  62. [70]

    AlDahoul, N. et al. Local receptive field-extreme learning machine based adult content detection. In 2019 IEEE International Confer- ence on Signal and Image Processing Applications (ICSIPA), 128–133 (IEEE, 2019)

    2019

  63. [71]

    Nian, F., Li, T., Wang, Y., Xu, M. & Wu, J. Pornographic image detectionutilizingdeepconvolutionalneuralnetworks. Neurocomputing 210, 283–293 (2016)

    2016

  64. [72]

    Transfer detection of yolo to focus cnn’s attention on nude regions for adult content detection.Symmetry 13, 26 (2020)

    AlDahoul, N.et al. Transfer detection of yolo to focus cnn’s attention on nude regions for adult content detection.Symmetry 13, 26 (2020)

    2020

  65. [73]

    S., Mansor, S., AlDahoul, N

    Lyn, H. S., Mansor, S., AlDahoul, N. & Karim, H. A. Convolutional neural network-based transfer learning and classification of visual con- tents for film censorship.Journal of Engineering Technology and Ap- plied Physics 2, 28–35 (2020)

    2020

  66. [74]

    L.et al.Deep active learning for pornography recognition using resnet

    Hor, S. L.et al.Deep active learning for pornography recognition using resnet. International Journal of Technology13, 1261–1270 (2022)

    2022

  67. [75]

    An evaluation of traditional and cnn-based feature descriptors for cartoon pornography detection.IEEE Access9, 39910– 39925 (2021)

    Aldahoul, N.et al. An evaluation of traditional and cnn-based feature descriptors for cartoon pornography detection.IEEE Access9, 39910– 39925 (2021)

    2021

  68. [76]

    AlDahoul, N., Karim, H.A., Wazir, A.S.B., Momo, M.A.&Abdullah, M. H. L. A comparative study of in-domain vs cross-domain learning for porn cartoon classification. In2021 IEEE International Conference on Signal and Image Processing Applications (ICSIPA), 60–65 (IEEE, 2021)

    2021

  69. [77]

    Momo, M. A.et al. Evaluation of convolution and attention networks for nudity and pornography detection in sketch images. In2023 IEEE Symposium on Computers & Informatics (ISCI), 7–12 (IEEE, 2023)

    2023

  70. [78]

    & Kong, A

    Huang, Y. & Kong, A. W. K. Using a cnn ensemble for detecting porno- graphicandupskirtimages. In 2016 IEEE 8th International Conference 50 on Biometrics Theory, Applications and Systems (BTAS), 1–7 (IEEE, 2016)

    2016

  71. [79]

    Wang, Y. & Li, W. Pornographic image recognition based on high and low level feature fusion with human body masking and attention. In Proceedings of the 2022 3rd International Conference on Control, Robotics and Intelligent System, 199–204 (2022)

    2022

  72. [80]

    & Joy, P

    Febin, I., Jayasree, K. & Joy, P. T. Violence detection in videos for an intelligent surveillance system using mobsift and movement filtering algorithm. Pattern Analysis and Applications23, 611–623 (2020)

    2020

  73. [81]

    S., Franqueira, V

    Nadeem, M. S., Franqueira, V. N., Kurugollu, F. & Zhai, X. Wvd: A new synthetic dataset for video-based violence detection. InArtificial Intelligence XXXVI: 39th SGAI International Conference on Artificial Intelligence, AI 2019, Cambridge, UK, December 17–19, 2019, Pro- ceedin...

    2019

  74. [82]

    & Tan, T

    Wang, D., Zhang, Z., Wang, W., Wang, L. & Tan, T. Baseline results for violence detection in still images. In2012 IEEE Ninth International Conference on Advanced Video and Signal-Based Surveillance, 54–57 (IEEE, 2012)

    2012

  75. [83]

    Real-time violence detection using cnn-lstm.arXiv preprint arXiv:2107.07578 (2021)

    Patel, M. Real-time violence detection using cnn-lstm.arXiv preprint arXiv:2107.07578 (2021)

    2021 arXiv

  76. [84]

    & Senan, N

    Ali, A. & Senan, N. Violence video classification performance using deep neural networks. In Recent Advances on Soft Computing and Data Mining: Proceedings of the Third International Conference on Soft Computing and Data Mining (SCDM 2018), Johor, Malaysia, February 06-07, 201...

    2018

  77. [85]

    & Singh, A

    Bagga, N., Singh, G., Balusamy, B. & Singh, A. S. Violence detec- tion in real life videos using convolutional neural network. In 2022 2nd International Conference on Advance Computing and Innovative Technologies in Engineering (ICACITE), 872–876 (IEEE, 2022)

    2022

  78. [86]

    Cheng, M., Cai, K. & Li, M. Rwf-2000: an open large scale video database for violence detection. In2020 25th International Conference on Pattern Recognition (ICPR), 4183–4190 (IEEE, 2021). 51

    2000

  79. [87]

    In2021 IEEE International Conference on Artificial Intelligence in Engineering and Technology (IICAIET), 1–6 (IEEE, 2021)

    AlDahoul, N.et al.Convolutionalneuralnetwork-longshorttermmem- ory based iot node for violence detection. In2021 IEEE International Conference on Artificial Intelligence in Engineering and Technology (IICAIET), 1–6 (IEEE, 2021)

    2021

  80. [88]

    Abdullah, M. S. N. B., Karim, H. A. & AlDahoul, N. A combination of light pre-trained convolutional neural networks and long short-term memory for real-time violence detection in videos.Methods 400, 4000 (2023)

    2023

  81. [89]

    Wang, H., Yang, L., Wu, X. & He, J. A review of bloody violence in video classification. In2017 International Conference on the Frontiers and Advances in Data Science (FADS), 86–91 (IEEE, 2017)

    2017

  82. [90]

    In2024 Asia Pacific Conference on Innovation in Technology (APCIT), 1–6 (IEEE, 2024)

    Doss, S.et al.Comparative analysis of news articles summarization us- ing llms. In2024 Asia Pacific Conference on Innovation in Technology (APCIT), 1–6 (IEEE, 2024)

    2024

  83. [91]

    Analysis of llms for educational question classification and generation.Computers and Education: Ar- tificial Intelligence 100298 (2024)

    Al Faraby, S., Romadhony, A.et al. Analysis of llms for educational question classification and generation.Computers and Education: Ar- tificial Intelligence 100298 (2024)

    2024

  84. [92]

    & Weber, I

    Nejjar, M., Zacharias, L., Stiehle, F. & Weber, I. Llms for science: Usage for code generation and data analysis. Journal of Software: Evolution and Processe2723 (2023)

    2023

  85. [93]

    & Huang, T.-H

    Huang, C.-Y., Wei, J. & Huang, T.-H. Generating educational ma- terials with different levels of readability using llms. arXiv preprint arXiv:2406.12787 (2024)

    2024 arXiv

  86. [94]

    Minaee, S. et al. Large language models: A survey. arXiv preprint arXiv:2402.06196 (2024)

    2024 arXiv

  87. [95]

    & Parapar, J

    Piot, P. & Parapar, J. Decoding hate: Exploring language models’ reactions to hate speech.arXiv preprint arXiv:2410.00775(2024)

    2024 arXiv

  88. [96]

    Yenduri, G.et al.Gpt (generative pre-trained transformer)–a compre- hensive review on enabling technologies, potential applications, emerg- ing challenges, and future directions.IEEE Access(2024). 52

    2024

  89. [97]

    Wu, T. et al. A brief overview of chatgpt: The history, status quo and potential future development.IEEE/CAA Journal of Automatica Sinica 10, 1122–1136 (2023)

    2023

  90. [98]

    Whenvisionmeetsreality: Exploring the clinical applicability of gpt-4 with vision (2024)

    Deng, J., Heybati, K.&Shammas-Toma, M. Whenvisionmeetsreality: Exploring the clinical applicability of gpt-4 with vision (2024)

    2024

  91. [99]

    Mu, Y.et al.Embodiedgpt: Vision-language pre-training via embodied chain of thought.Advances in Neural Information Processing Systems 36 (2024)

    2024

  92. [100]

    & Hoiem, D

    Gupta, T., Kamath, A., Kembhavi, A. & Hoiem, D. Towards general purpose vision systems: An end-to-end task-agnostic vision-language architecture. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, 16399–16409 (2022)

    2022

  93. [101]

    Vqa: Visual question answering

    Antol, S.et al. Vqa: Visual question answering. InProceedings of the IEEE international conference on computer vision, 2425–2433 (2015)

    2015

  94. [102]

    Gemini 1.5 technical report

    Gemini Team, G. Gemini 1.5 technical report. https: //storage.googleapis.com/deepmind-media/gemini/gemini_ v1_5_report.pdf (2024)

    2024

  95. [103]

    https://blog.google/technology/ai/ google-gemini-next-generation-model-february-2024/ #architecture (2024)

    Introducing gemini 1.5, google’s next-generation ai model. https://blog.google/technology/ai/ google-gemini-next-generation-model-february-2024/ #architecture (2024)

    2024

  96. [104]

    https://openai.com/index/hello-gpt-4o// (2024)

    Hello gpt-4o. https://openai.com/index/hello-gpt-4o// (2024)

    2024

  97. [105]

    Gpt-4o (2024)

    OpenAI. Gpt-4o (2024). URLhttps://platform.openai.com/docs/ models/gpt-4o

    2024

  98. [106]

    tdavidson/hate_speech_offensive

    Face, H. tdavidson/hate_speech_offensive. URL https: //huggingface.co/datasets/tdavidson/hate_speech_offensive/ tree/main/data

  99. [107]

    Gender-based violence tweet classification

    Dutta, G. Gender-based violence tweet classification. URL https://www.kaggle.com/datasets/gauravduttakiit/ gender-based-violence-tweet-classification?select=Train. csv. 53

  100. [108]

    valurank/adult-content-dataset

    Face, H. valurank/adult-content-dataset. https://huggingface.co/ datasets/valurank/Adult-content-dataset

  101. [109]

    Hou, Y. et al. Bridging language and items for retrieval and recom- mendation. arXiv preprint arXiv:2403.03952(2024)

    2024 arXiv

  102. [110]

    https://aws.amazon.com/blogs/machine-learning/ how-amazon-shopping-uses-amazon-rekognition-content-moderation-to-review-harmful-images-in-product-reviews/

    How amazon shopping uses amazon rekognition con- tent moderation to review harmful images in product re- views. https://aws.amazon.com/blogs/machine-learning/ how-amazon-shopping-uses-amazon-rekognition-content-moderation-to-review-harmful-images-in-product-reviews/

  103. [111]

    com/transcribe/toxicity-detection/

    Why amazon transcribe toxicity detection? https://aws.amazon. com/transcribe/toxicity-detection/

  104. [112]

    In 2019 ninth international conference on intelligent computing and information systems (ICICIS), 80–85 (IEEE, 2019)

    Soliman, M.M.et al.Violencerecognitionfromvideosusingdeeplearn- ing techniques. In 2019 ninth international conference on intelligent computing and information systems (ICICIS), 80–85 (IEEE, 2019)

    2019

  105. [113]

    meta-llama/llama-guard-3-8b

    Face, H. meta-llama/llama-guard-3-8b. https://huggingface.co/ meta-llama/Llama-Guard-3-8B

  106. [114]

    meta-llama/llama-guard-3-11b-vision

    Face, H. meta-llama/llama-guard-3-11b-vision. https: //huggingface.co/meta-llama/Llama-Guard-3-11B-Vision

  107. [115]

    meta-llama/llama-3.1-8b-instruct

    Face, H. meta-llama/llama-3.1-8b-instruct. https://huggingface. co/meta-llama/Llama-3.1-8B-Instruct

  108. [116]

    meta-llama/llama-3.2-11b-vision-instruct

    Face, H. meta-llama/llama-3.2-11b-vision-instruct. https:// huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct

  109. [117]

    Gemini, google’s most capable model, is now avail- able on vertex ai

    Cloud, G. Gemini, google’s most capable model, is now avail- able on vertex ai. https://cloud.google.com/blog/products/ ai-machine-learning/gemini-support-on-vertex-ai (2023)

    2023

  110. [118]

    safety-settings

    Gemini Team, G. safety-settings. https://ai.google.dev/ gemini-api/docs/safety-settings (2024). 54

    2024

Pith tools

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