Pith. sign in

REVIEW 4 major objections 5 minor 57 references

Scout: Leveraging Large Language Models for Rapid Digital Evidence Discovery

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Scout presents a prototype framework in which large language models pre-screen seized digital evidence and rank potentially relevant files before manual analysis.

desk verdict Scout is a sensible prototype idea with an honest limitations section, but its central claim is never tested: no accuracy metrics, no baseline, no code, and the Results section is screenshots plus an unfinished sentence. read the letter →

arxiv 2507.18478 v1 pith:THPV5LTH submitted 2025-07-24 cs.CR

Shariq Murtuza This is my paper
classification cs.CR
keywords digitalforensicslargelanguagemodelsevidencetriagemultimodalprioritizationnetworkpacketanalysisEnrondatasetforensicpreprocessing
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

Scout is a prototype framework for digital forensics that uses large language models to look through seized digital evidence and tell an investigator which files deserve attention first. The paper's aim is to show that general-purpose text models and multimodal models can serve as a pre-analysis layer, processing network packets, emails, office documents, audio, images, and video, and flagging artifacts that are potentially relevant to an ongoing investigation. The value, if the claim holds, is that an examiner facing gigabytes of data can start from a shortlist suggested by the models rather than guessing where to begin. Scout is positioned strictly as a triage aid: it never extracts or presents evidence itself, and the paper explicitly withholds quantitative accuracy figures because the models are probabilistic.

What carries the argument

The carrying mechanism is Scout's evidence-type routing pipeline. Each seized file is first normalized into a model-friendly form: network packet captures are fed packet by packet, office documents are converted to structured content by Docling, and audio files are transcribed by Whisper before the text goes to a language model. The prepared content is then passed to a foundation model along with a prompt that can carry investigation context, and the model returns observations, summaries, and red flags that the investigator can use to prioritize manual review. Scout runs read-only and can be deployed offline, and its per-file-type design means new evidence formats can be added as plugins.

What would settle it

Take a corpus of seized files with known ground-truth relevant artifacts, run Scout with the same prompts used in the paper, and check whether the known-relevant files appear at the top of the suggested processing order; if the ranking is no better than random, the central prioritization claim fails.

Watch

Extended reading notes

Core claim

In the paper's own terms, the discovery is that a prompt-driven pipeline built around foundation models can identify, summarize, and contextualize forensic artifacts well enough to recommend a processing order for evidence files. On a sample network packet capture, the framework noticed repeated DNS requests and ICMP errors; on Enron emails it summarized conversations and flagged meetings and purchases; on an office document it spotted that the creation date had been set in the future; and on images and videos it described scenes and raised case-related observations. The authors state that Scout was able to identify the evidence files that were of potential interest to the investigator, and that it does so by choosing text models for textual evidence and multimodal models for multimedia evidence, with the investigator able to rerun the evidence through different models or prompts.

Load-bearing premise

Scout's usefulness rests on the assumption that the files the models flag are genuinely the ones an investigator would find most relevant, and the paper provides no accuracy measurement that would test that match.

Editorial extensions

If this is right

  • Investigators can begin a case by reviewing a model-generated shortlist, which may reduce the time spent sifting large disk images, packet captures, and media collections.
  • Evidence stays admissible in principle because Scout only points at files; extraction and examination remain with the investigator using approved methods.
  • Because models can be swapped and rerun, the same evidence can be examined from multiple interpretive angles without altering the underlying files.
  • A fully offline deployment lets agencies process sensitive seized material without sending it to external services.
  • Extending Scout to new file types becomes a plugin-writing exercise, so the approach can keep pace with new devices and formats.

Reading between the lines

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

  • If model-driven triage proves reliable, the practical bottleneck in digital forensics shifts from discovery to verification: resources concentrate on confirming flagged artifacts rather than on finding them.
  • The author's own warning about false negatives implies that the framework should be measured by ranking quality, not by whether it ever misses, and that any deployment should pair model flags with exhaustive enumeration.
  • A testable extension would be to run Scout on a standardized corpus with known relevant artifacts and compare the position of true positives against random ordering.
  • The approach could be combined with keyword and hash-based tools, since language models and signature tools catch different kinds of leads: contextual anomalies versus exact matches.
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

4 major / 5 minor

Summary. The paper presents Scout, a prototype framework that uses large language models and multimodal models to pre-analyze and prioritize digital evidence files (network packet captures, emails, office documents, audio, images, videos) for forensic investigators. The claimed contribution is that Scout suggests an order for processing files so that relevant evidence is found faster. The paper describes the architecture, model choices (e.g., Llama 3.3, Hermes 3, Qwen2-VL, Whisper), and shows qualitative screenshots of model outputs on a small set of example files. The authors explicitly state that they withhold accuracy metrics due to the probabilistic nature of foundation models, and they acknowledge a large possibility of false negatives. No baseline comparison, quantitative evaluation, or ground-truth validation is provided.

Significance. If validated, a tool that reliably prioritizes digital evidence for forensic triage would be a practically useful contribution, and the proposed design has plausible strengths: it operates offline, in read-only mode to preserve evidence integrity, and is plugin-extensible across file types. However, the manuscript's central claim—that Scout enables rapid evidence discovery—is entirely unsupported by the reported results. The paper ships no machine-checked proofs, no reproducible evaluation pipeline, and no falsifiable quantitative predictions; it instead withholds accuracy data and presents only successful examples. The significance of the work therefore remains speculative, and in its current form the paper is a prototype description rather than a validated research contribution.

major comments (4)
  1. [Section 4, opening paragraph] The central claim of the paper is that Scout helps investigators find relevant evidence faster, but Section 4 explicitly states: "We decided to hold the release of Scout's accuracy on complete datasets due to the probabilistic nature of foundation models." This is a load-bearing omission: no precision, recall, F1, time-to-evidence, or comparison against random ordering, keyword search, or existing triage tools is reported. The screenshots in Figures 3–14 are anecdotal and do not substantiate the prioritization quality that the contribution promises.
  2. [Section 5] The discussion concedes "a large possibility of false negatives" and states that "All the remaining files are must also be analyzed by the forensic investigator." Together with the withheld accuracy metrics, this means a missed file leaves the full workload intact, so the proposed benefit of reduced analysis time is never demonstrated. The paper provides no measurement of how often Scout's ranking improves over the original file order, which is the core requirement for the stated use case.
  3. [Section 4.4] The audio analysis subsection ends with the incomplete sentence "This XXX". This is not a minor typo; it indicates that the audio analysis result is unfinished. Given that audio is one of the six evidence modalities claimed as supported, the incompleteness undermines the completeness of the presented evaluation.
  4. [Figures 11–14] Figures 11 and 13 are labeled "Sample video" but the corresponding output captions (Figures 12 and 14) say "Information gathered from the previous image." This mislabeling, along with the fact that only successful outputs are shown with no failure cases or error analysis, suggests the reported examples may be post-hoc selections. The paper does not describe the sampling of test files or the selection criteria for the displayed outputs, making it impossible to assess representativeness.
minor comments (5)
  1. [Section 4.2] The sentence "Scout was able was to summarize the emails" contains a duplicated verb and should be corrected.
  2. [Section 3.5] The phrase "since the larger and clearer the image is, the more accurate the analysis is" is informal; consider rephrasing for clarity, and specify the resolution/downscaling trade-off with quantitative details if available.
  3. [Section 2.2] The term "foundational models" is used inconsistently with "foundation models" in the abstract; unify the terminology.
  4. [References] Several references are incomplete or informal (e.g., [28], [42], [44], [54]), lacking page numbers, publisher information, or access dates; the reference list should be formatted to the journal's standards.
  5. [Throughout] The paper contains numerous typographical and grammatical errors (e.g., "o iff ce"), inconsistent capitalization, and informal phrasings that require thorough proofreading.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: Scout reports raw LLM outputs with no fitted parameters or self-referential equations; withheld accuracy is a validation gap, not circularity.

full rationale

The paper's central claim is that Scout, a prototype framework, uses pretrained foundation models to flag potentially interesting evidence files. There are no equations, no fitted parameters, and no quantity is predicted from a separately fitted input. The outputs are direct responses from external models (e.g., Llama 3.3, Hermes 3, Qwen2-VL, Whisper, Docling), so no step reduces to its own inputs by construction. The only self-citation, [14], appears in a general list of digital forensics process references and is not load-bearing for the framework's behavior. Section 4 explicitly withholds accuracy metrics due to the probabilistic nature of foundation models, and Section 5 acknowledges false negatives; these are empirical soundness and validation concerns, not circularity. Because the reported results are demonstrations rather than derived predictions, the derivation chain, such as it is, is self-contained and does not invert inputs into outputs.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

No new theoretical quantities or entities are introduced. The framework uses existing pretrained models as black boxes, so there are no fitted parameters. The central assumptions are about the reliability of model outputs and the preservation of evidentiary context.

assumptions (3)
  • domain assumption Foundation model outputs are reliable enough to prioritize evidence without systematic validation
    Section 4 withholds accuracy metrics and Section 5 acknowledges false negatives; the entire framework depends on this untested premise.
  • domain assumption Processing each packet or file independently preserves the information needed for relevance judgment
    Section 3.1 describes sending packets one by one; no evidence is provided that this avoids losing cross-packet context.
  • domain assumption The models operate in a read-only manner that does not contaminate evidence
    Section 5 states read-only mode, but no integrity verification or hash checks are shown to support this claim.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Scout: Leveraging Large Language Models for Rapid Digital Evidence Discovery." pith.science (2026). https://pith.science/paper/THPV5LTH

@misc{pith2026250718478,
  author       = {Pith},
  title        = {Pith review of: Scout: Leveraging Large Language Models for Rapid Digital Evidence Discovery},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/THPV5LTH}},
  note         = {Machine review of arXiv:2507.18478}
}
read the original abstract

Recent technological advancements and the prevalence of technology in day to day activities have caused a major increase in the likelihood of the involvement of digital evidence in more and more legal investigations. Consumer-grade hardware is growing more powerful, with expanding memory and storage sizes and enhanced processor capabilities. Forensics investigators often have to sift through gigabytes of data during an ongoing investigation making the process tedious. Memory forensics, disk analysis all are well supported by state of the art tools that significantly lower the effort required to be put in by a forensic investigator by providing string searches, analyzing images file etc. During the course of the investigation a lot of false positives are identified that need to be lowered. This work presents Scout, a digital forensics framework that performs preliminary evidence processing and prioritizing using large language models. Scout deploys foundational language models to identify relevant artifacts from a large number of potential evidence files (disk images, captured network packets, memory dumps etc.) which would have taken longer to get identified. Scout employs text based large language models can easily process files with textual information. For the forensic analysis of multimedia files like audio, image, video, office documents etc. multimodal models are employed by Scout. Scout was able to identify and realize the evidence file that were of potential interest for the investigator.

Figures

Figures reproduced from arXiv: 2507.18478 by the authors.

Figure 1
Figure 1. Casey’s Traditional Digital Forensic Model [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Scout Framework on the work done by the authors and aims to extend their work. 3 Experimental Methodology This section details the steps that were taken for processing of evidence and analysis using the foundational models to identify files of potential interest. Different seized evidence files requires different models for processing, evidence may have network packets, texts (emails, office documents) etc. are proc… view at source ↗
Figure 3
Figure 3. Analysis of network packets from PCAP files. Enron dataset. The results obtained reflected the typical understanding of a large language model. The amount of data that can be extracted from the Enron email dataset depends highly upon the prompt and the model finetuning. Scout was able was to summarize the emails and raise observations from the chats such as meetings and purchases. Two sample runs shown in figures 4 … view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Sample email Analysis from Enron Dataset [PITH_FULL_IMAGE:figures/full_fig_p027_4.png]
Figure 5
Figure 5. Figure 5: Sample email Analysis from Enron Dataset [PITH_FULL_IMAGE:figures/full_fig_p027_5.png]
Figure 6
Figure 6. Figure 6: Analysis of a Tampered office document with creation date set later than the last modified date [PITH_FULL_IMAGE:figures/full_fig_p028_6.png]
Figure 7
Figure 7. Figure 7: Sam￾ple image taken from Digital Cor￾pora [56] [PITH_FULL_IMAGE:figures/full_fig_p028_7.png]
Figure 9
Figure 9. Figure 9: Sample image taken from Digital Corpora [56] [PITH_FULL_IMAGE:figures/full_fig_p029_9.png]
Figure 12
Figure 12. Figure 12: Information gathered from the previous image. The [PITH_FULL_IMAGE:figures/full_fig_p029_12.png]
Figure 14
Figure 14. Figure 14: Information gathered from the previous image. The [PITH_FULL_IMAGE:figures/full_fig_p030_14.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

57 extracted references · 42 canonical work pages

  1. [1]

    The digital crime tsunami

    Gogolin, G., 2010. The digital crime tsunami. Digital investigation, 7(1-2), pp.3-8

    work page 2010

  2. [2]

    and Choo, K.K.R., 2014

    Quick, D. and Choo, K.K.R., 2014. Impacts of increasing volume of digital forensic data: A survey and future research challenges. Digital Investigation, 11(4), pp.273-294

    work page 2014

  3. [3]

    and Martell, R., 2013

    Roussev, V., Quates, C. and Martell, R., 2013. Real-time digital forensics and triage. Digital Investigation, 10(2), pp.158-167

    work page 2013

  4. [4]

    and Browne, A., 2013

    Shaw, A. and Browne, A., 2013. A practical and robust approach to coping with large volumes of data submitted for digital forensic examination. Digital Investigation, 10(2), pp.116-128. 19

    work page 2013

  5. [5]

    and Mian, A., 2023

    Naveed, H., Khan, A.U., Qiu, S., Saqib, M., Anwar, S., Usman, M., Akhtar, N., Barnes, N. and Mian, A., 2023. A comprehensive overview of large language models. arXiv preprint arXiv:2307.06435

    arXiv 2023

  6. [6]

    Common Crawl: Data Collection and Use Cases for NLP

    Nagel, S., 2023. Common Crawl: Data Collection and Use Cases for NLP. HPLT & NLPL Winter School on Large-Scale Language Modeling and Neural Machine Transla- tion with Web Data, February, 6

    work page 2023

  7. [7]

    and Khan, F.S., 2023

    Awais, M., Naseer, M., Khan, S., Anwer, R.M., Cholakkal, H., Shah, M., Yang, M.H. and Khan, F.S., 2023. Foundational models de fining a new era in vision: A survey and outlook. arXiv preprint arXiv:2307.13721

    arXiv 2023

  8. [8]

    and Chacko, V., 2023

    Chacko, N. and Chacko, V., 2023. Paradigm shift presented by large language models (llm) in deep learning. ADV ANCES IN EMERGING COMPUTING TECHNOLOGIES, 40

    work page 2023

Show all 57 references
  1. [9]

    and Polosukhin, I., 2017

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L. and Polosukhin, I., 2017. Attention Is All You Need.(Nips), 2017. arXiv preprint arXiv:1706.03762, 10, p.S0140525X16001837

    2017 arXiv

  2. [10]

    BERT: a review of applications in natural language processing and understanding

    Koroteev, M.V., 2021. BERT: a review of applications in natural language processing and understanding. arXiv preprint arXiv:2103.11943

    2021 arXiv

  3. [11]

    and Agarwal, S., 2020

    Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J.D., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A. and Agarwal, S., 2020. Language models are few- shot learners. Advances in neural information processing systems, 33, pp.1877-1901

    2020

  4. [12]

    Pollitt, Mark. ”A history of digital forensics.” Advances in Digital Forensics VI: Sixth IFIP WG 11.9 International Conference on Digital Forensics, Hong Kong, China, Jan- uary 4-6, 2010, Revised Selected Papers 6. Springer Berlin Heidelberg, 2010. 20

    2010

  5. [13]

    ”Digital forensics research: The next 10 years.” digital investigation 7 (2010): S64-S73

    Gar finkel, Simson L. ”Digital forensics research: The next 10 years.” digital investigation 7 (2010): S64-S73

    2010

  6. [14]

    ”A tool for extracting static and volatile forensic artifacts of windows 8

    Murtuza, Shariq, Robin Verma, Jayaprakash Govindaraj, and Gaurav Gupta. ”A tool for extracting static and volatile forensic artifacts of windows 8. x apps.” In IFIP In- ternational Conference on Digital Forensics, pp. 305-320. Cham: Springer International Publishing, 2015

    2015

  7. [15]

    Ruan, Keyun, et al. ”Cloud forensics.” Advances in Digital Forensics VII: 7th IFIP WG 11.9 International Conference on Digital Forensics, Orlando, FL, USA, January 31–February 2, 2011, Revised Selected Papers 7. Springer Berlin Heidelberg, 2011

    2011

  8. [16]

    ”Cloud forensics: Technical chal- lenges, solutions and comparative analysis.” Digital investigation 13 (2015): 38-57

    Pichan, Ameer, Mihai Lazarescu, and Sie Teng Soh. ”Cloud forensics: Technical chal- lenges, solutions and comparative analysis.” Digital investigation 13 (2015): 38-57

    2015

  9. [17]

    ”Mobile forensics: advances, challenges, and research oppor- tunities.” IEEE Security & Privacy 15.6 (2017): 42-51

    Chernyshev, Maxim, et al. ”Mobile forensics: advances, challenges, and research oppor- tunities.” IEEE Security & Privacy 15.6 (2017): 42-51

    2017

  10. [18]

    ”Current and future trends in mobile device forensics: A survey.” ACM Computing Surveys (CSUR) 51.3 (2018): 1-31

    Barmpatsalou, Konstantia, et al. ”Current and future trends in mobile device forensics: A survey.” ACM Computing Surveys (CSUR) 51.3 (2018): 1-31

    2018

  11. [19]

    ”A survey on the internet of things (IoT) forensics: challenges, approaches, and open issues.” IEEE Communications Surveys & Tutorials 22.2 (2020): 1191-1221

    Stoyanova, Maria, et al. ”A survey on the internet of things (IoT) forensics: challenges, approaches, and open issues.” IEEE Communications Surveys & Tutorials 22.2 (2020): 1191-1221

    2020

  12. [20]

    ”Internet of things forensics: A review.” Internet of Things 11 (2020): 100220

    Atlam, Hany F., et al. ”Internet of things forensics: A review.” Internet of Things 11 (2020): 100220

    2020

  13. [21]

    ”Internet of things forensics: Recent advances, taxonomy, require- ments, and open challenges.” Future Generation Computer Systems 92 (2019): 265-275

    Yaqoob, Ibrar, et al. ”Internet of things forensics: Recent advances, taxonomy, require- ments, and open challenges.” Future Generation Computer Systems 92 (2019): 265-275. 21

    2019

  14. [22]

    ”The impact of automation and arti fi- cial intelligence on digital forensics.” Wiley Interdisciplinary Reviews: Forensic Science 3.6 (2021): e1418

    Jarrett, Aaron, and Kim-Kwang Raymond Choo. ”The impact of automation and arti fi- cial intelligence on digital forensics.” Wiley Interdisciplinary Reviews: Forensic Science 3.6 (2021): e1418

    2021

  15. [23]

    ”Explainable arti ficial intelligence for digital forensics.” Wiley Interdisciplinary Reviews: Forensic Science 4.2 (2022): e1434

    Hall, Stuart W., Amin Sakzad, and Kim-Kwang Raymond Choo. ”Explainable arti ficial intelligence for digital forensics.” Wiley Interdisciplinary Reviews: Forensic Science 4.2 (2022): e1434

    2022

  16. [24]

    Al-Dhaqm, S

    A. Al-Dhaqm, S. A. Razak, R. A. Ikuesan, V. R. Kebande, K. Siddique, A Review of Mobile Forensic Investigation Process Models, IEEE Access 8 (2020) 173359–173375. doi:10.1109/ACCESS.2020.3014615

    2020

  17. [25]

    and Alsewari, A.A., 2021

    Al-Dhaqm, A., Ikuesan, R.A., Kebande, V.R., Abd Razak, S., Grispos, G., Choo, K.K.R., Al-Rimy, B.A.S. and Alsewari, A.A., 2021. Digital forensics subdomains: the state of the art and future directions. IEEE Access, 9, pp.152476-152502

    2021

  18. [26]

    and Gadekallu, T.R., 2022

    Javed, A.R., Ahmed, W., Alazab, M., Jalil, Z., Kifayat, K. and Gadekallu, T.R., 2022. A comprehensive survey on computer forensics: State-of-the-art, tools, techniques, chal- lenges, and future directions. IEEE Access, 10, pp.11065-11089

    2022

  19. [27]

    Handbook of digital forensics and investigation

    Casey, E., 2009. Handbook of digital forensics and investigation. Academic Press

    2009

  20. [28]

    and Nisbet, A., 2016

    Lennox-Steele, A. and Nisbet, A., 2016. A forensic examination of several mobile device Faraday bags & materials to test their e ffectiveness

    2016

  21. [29]

    and Khan, I.U., Arti ficial General Intelligence (AGI) Secu- rity

    El Hajjami, S., Kaushik, K. and Khan, I.U., Arti ficial General Intelligence (AGI) Secu- rity

  22. [30]

    and Kim, A., 2023

    Chew, R., Bollenbacher, J., Wenger, M., Speer, J. and Kim, A., 2023. LLM-assisted con- tent analysis: Using large language models to support deductive coding. arXiv preprint arXiv:2306.14924. 22

    2023 arXiv

  23. [31]

    and Ku, L.W., 2023

    Dai, S.C., Xiong, A. and Ku, L.W., 2023. LLM-in-the-loop: Leveraging large language model for thematic analysis. arXiv preprint arXiv:2310.15100

    2023 arXiv

  24. [32]

    and Zhang, J., 2023

    Liu, S.C., Wang, S., Lin, W., Hsiung, C.W., Hsieh, Y.C., Cheng, Y.P., Luo, S.H., Chang, T. and Zhang, J., 2023. Jarvix: A llm no code platform for tabular data analysis and optimization. arXiv preprint arXiv:2312.02213

    2023 arXiv

  25. [33]

    and Avila, R., 2023

    Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F.L., Almeida, D., Altenschmidt, J., Altman, S., Anadkat, S. and Avila, R., 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774

    2023 arXiv

  26. [34]

    and Rodriguez, A., 2023

    Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M.A., Lacroix, T., Rozi` ere, B., Goyal, N., Hambro, E., Azhar, F. and Rodriguez, A., 2023. Llama: Open and effi cient foundation language models. arXiv preprint arXiv:2302.13971

    2023 arXiv

  27. [35]

    and Silver, D., 2023

    Team, G., Anil, R., Borgeaud, S., Alayrac, J.B., Yu, J., Soricut, R., Schalkwyk, J., Dai, A.M., Hauth, A., Millican, K. and Silver, D., 2023. Gemini: a family of highly capable multimodal models. arXiv preprint arXiv:2312.11805

    2023 arXiv

  28. [36]

    and Hui, B., 2023

    Bai, J., Bai, S., Chu, Y., Cui, Z., Dang, K., Deng, X., Fan, Y., Ge, W., Han, Y., Huang, F. and Hui, B., 2023. Qwen technical report. arXiv preprint arXiv:2309.16609

    2023 arXiv

  29. [37]

    and Zhou, J., 2023

    Bai, J., Bai, S., Yang, S., Wang, S., Tan, S., Wang, P., Lin, J., Zhou, C. and Zhou, J., 2023. Qwen-vl: A frontier large vision-language model with versatile abilities. arXiv preprint arXiv:2308.12966

    2023 arXiv

  30. [38]

    OpenAI. (2024). ChatGPT [Large language model]. https://chat.openai.com

    2024

  31. [39]

    and Chiriatti, M., 2020

    Floridi, L. and Chiriatti, M., 2020. GPT-3: Its nature, scope, limits, and consequences. Minds and Machines, 30, pp.681-694

    2020

  32. [40]

    and Peng, H., 2024

    Zhou, C., Li, Q., Li, C., Yu, J., Liu, Y., Wang, G., Zhang, K., Ji, C., Yan, Q., He, L. and Peng, H., 2024. A comprehensive survey on pretrained foundation models: A history 23 from bert to chatgpt. International Journal of Machine Learning and Cybernetics, pp.1- 65

    2024

  33. [41]

    and Batool, L., 2024

    Shahriar, S., Lund, B.D., Mannuru, N.R., Arshad, M.A., Hayawi, K., Bevara, R.V.K., Mannuru, A. and Batool, L., 2024. Putting gpt-4o to the sword: A comprehensive evaluation of language, vision, speech, and multimodal pro ficiency. Applied Sciences, 14(17), p.7782

    2024

  34. [42]

    and Tselikas, N.D., 2023

    Roumeliotis, K.I. and Tselikas, N.D., 2023. Chatgpt and open-ai models: A preliminary review. Future Internet, 15(6), p.192

    2023

  35. [43]

    and Goyal, A., 2024

    Dubey, A., Jauhri, A., Pandey, A., Kadian, A., Al-Dahle, A., Letman, A., Mathur, A., Schelten, A., Yang, A., Fan, A. and Goyal, A., 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783

    2024 arXiv

  36. [44]

    Roque, L., The Evolution of Llama: From Llama 1 to Llama 3.1 A Comprehensive Guide to the Advancements and Innovations in the Family of Llama Models from Meta AI

  37. [45]

    and Mishra, L., 2024

    Auer, C., Lysak, M., Nassar, A., Dol fi, M., Livathinos, N., Vagenas, P., Ramis, C.B., Omenetti, M., Lindlbauer, F., Dinkla, K. and Mishra, L., 2024. Docling Technical Report. arXiv preprint arXiv:2408.09869

    2024 arXiv

  38. [46]

    and Staar, P., 2022, August

    P fitzmann, B., Auer, C., Dol fi, M., Nassar, A.S. and Staar, P., 2022, August. Doclaynet: A large human-annotated dataset for document-layout segmentation. In Proceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining (pp. 3743- 3751)

    2022

  39. [47]

    and Staar, P., 2022

    Nassar, A., Livathinos, N., Lysak, M. and Staar, P., 2022. Tableformer: Table struc- ture understanding with transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 4614-4623). 24

    2022

  40. [48]

    and Scanlon, M., 2024

    Wickramasekara, A., Breitinger, F. and Scanlon, M., 2024. Exploring the Potential of Large Language Models for Improving Digital Forensic Investigation E ffi ciency. arXiv preprint arXiv:2402.19366

    2024 arXiv

  41. [49]

    and Sheppard, J., 2023

    Scanlon, M., Breitinger, F., Hargreaves, C., Hilgert, J.N. and Sheppard, J., 2023. Chat- GPT for digital forensic investigation: The good, the bad, and the unknown. Forensic Science International: Digital Investigation, 46, p.301609

    2023

  42. [50]

    and Liu, T., 2023

    Huang, L., Yu, W., Ma, W., Zhong, W., Feng, Z., Wang, H., Chen, Q., Peng, W., Feng, X., Qin, B. and Liu, T., 2023. A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions. ACM Transactions on Information Systems

    2023

  43. [51]

    and Guang, C., 2024

    Teknium, R., Quesnelle, J. and Guang, C., 2024. Hermes 3 technical report. arXiv preprint arXiv:2408.11857

    2024 arXiv

  44. [52]

    and Sutskever, I., 2023, July

    Radford, A., Kim, J.W., Xu, T., Brockman, G., McLeavey, C. and Sutskever, I., 2023, July. Robust speech recognition via large-scale weak supervision. In International con- ference on machine learning (pp. 28492-28518). PMLR

    2023

  45. [53]

    and Lee, Y.J., 2024

    Liu, H., Li, C., Li, Y. and Lee, Y.J., 2024. Improved baselines with visual instruction tuning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 26296-26306)

    2024

  46. [54]

    huggingface.co/Intel/llava-llama-3-8b

    Intel/llava-llama-3-8b · Hugging Face. huggingface.co/Intel/llava-llama-3-8b

  47. [55]

    and Fan, Y., 2024

    Wang, P., Bai, S., Tan, S., Wang, S., Fan, Z., Bai, J., Chen, K., Liu, X., Wang, J., Ge, W. and Fan, Y., 2024. Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution. arXiv preprint arXiv:2409.12191

    2024 arXiv

  48. [56]

    Gar finkel, Farrell, Roussev and Dinolt, Bringing Science to Digital Forensics with Stan- dardized Forensic Corpora, DFR WS 2009, Montreal, Canada 25

    2009

  49. [57]

    LibriVox — free public domain audiobooks. (n.d.). https://librivox.org/ 26 Figure 4: Sample email Analysis from Enron Dataset (Names blurred). Figure 5: Sample email Analysis from Enron Dataset (Names blurred). 27 Figure 6: Analysis of a Tampered o ffi ce document with creation ...

Pith tools

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