Pith. sign in

REVIEW 3 major objections 5 minor 71 references

A single system can turn one research paper into a consistent suite of posters, slides, videos and project pages by planning from shared evidence and repairing cross-format conflicts.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-12 09:29 UTC pith:3BBP2WIK

load-bearing objection Solid systems paper that unifies multi-format paper-to-suite generation with a real verify-and-repair loop and a large paired benchmark; main risk is Gemini-as-judge circularity, not the core design. the 3 major comments →

arxiv 2607.02590 v1 pith:3BBP2WIK submitted 2026-07-01 cs.SE

OmniPresent: Generating Coherent Presentation Suites from Scientific Papers

Qianli Ma , Jipeng Xiao , Siyu Wang , Zhiheng Tian , Wangyu Feng , Shibo Wang , Chang Guo , Shuochen Chang
show 2 more authors
Qingyang Liu Zhipeng Zhang
This is my paper
classification cs.SE
keywords presentation suite generationscientific disseminationcross-format consistencymulti-agent planningverify-and-repairHTML intermediate representationOmniPreBenchVLM evaluation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Generating posters, slides, videos and project pages from a finished paper is still mostly manual and error-prone. Current automated tools handle each format separately, so numbers, claims and terminology routinely drift out of alignment across the suite. This paper formalizes the whole suite as one generation task and shows that a shared knowledge stream of evidence-linked claims, plus a verify-and-repair loop that catches conflicts and coverage gaps, can keep every artifact faithful to the source and consistent with the others. Everything is emitted as renderable HTML so the same content plan can drive static layouts and dynamic flows. The result is higher factual accuracy and better visual quality than strong specialized baselines, together with a new public benchmark of more than a thousand papers paired with real multi-format artifacts.

Core claim

Unified presentation-suite generation is both feasible and superior to isolated pipelines when content is first distilled into a single evidence-grounded knowledge stream, planned under format-specific constraints, and then iteratively checked and repaired for cross-artifact conflicts and missing coverage before any HTML is rendered.

What carries the argument

The cross-artifact verify-and-repair loop that sits on top of a shared knowledge stream of (claim, evidence, type) triples and a universal renderable-HTML intermediate representation; the loop extracts claims from every plan, flags unsupported or contradictory statements and salience gaps, then rewrites the plans until the suite converges.

Load-bearing premise

That the automated VLM/LLM judges used for FactScore, Q&A accuracy and visual rubrics are faithful enough proxies for true scientific correctness and human visual quality that large metric gains can be read as genuine suite-level superiority.

What would settle it

A blinded human study in which domain experts, scoring only cross-format factual consistency and coverage, systematically prefer the specialized single-format baselines over OmniPresent, or in which the reported FactScore gains reverse when the same artifacts are re-scored by an independent model family never used in training or evaluation.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Researchers can produce a complete, fact-aligned dissemination package from a single paper with far less manual reconciliation of numbers and claims.
  • Suite-level consistency becomes an enforceable system property rather than an after-the-fact authoring chore.
  • Weaker backbone models can still generate high-quality multi-format suites when constrained by centralized planning and verification.
  • A public multi-format benchmark now exists so future methods can be compared on faithfulness and cross-artifact alignment, not just single-format aesthetics.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same shared-stream-plus-repair pattern could be applied to other multi-view scientific documents such as grant summaries, press releases and teaching notes that must stay consistent with a core paper.
  • Because the intermediate form is editable HTML, the architecture naturally supports interactive human-in-the-loop polishing that preserves cross-format guarantees after each edit.
  • Extending the benchmark and the pipeline beyond AI conference papers would test whether domain-specific visual conventions (chemical structures, clinical imaging) still admit the same centralized planning strategy.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper formalizes unified presentation-suite generation (poster, slides, video, project page) from a single scientific paper and introduces OmniPresent, a multi-agent Plan-Verify-Render system that uses a shared Knowledge Stream K of evidence-grounded units (c,e,t), format-specific planners, a cross-artifact verify-and-repair loop (Algorithm 1, Eqs. 5–6) enforcing salience containment (Eq. 1), and adaptive HTML rendering. It releases OmniPreBench (1,037 paired multi-format papers, 50-paper stratified test set) and a VLM/LLM-as-judge protocol (FactScore via Eq. 7, QA, readability, visual accuracy/aesthetics/layout, interaction success). Controlled experiments holding backbones fixed, ablations of the verification loop (Tab. 3), inclusion/conflict metrics (Tabs. 4, 6), a 10-paper non-AI set (Tab. 5), and a 13-person visual preference study (Tab. 7) show OmniPresent outperforming specialized systems (EvoPresent, Paper2Poster, AutoPage, PaperX) and end-to-end models on content faithfulness and visual quality.

Significance. If the reported gains hold under independent evaluation, the work is a clear advance for automated scientific communication: it is the first system that jointly produces a coherent four-format suite rather than isolated artifacts, supplies a large paired benchmark that the community currently lacks, and demonstrates that centralized evidence indexing plus a verify-and-repair loop measurably reduces cross-format drift. The HTML intermediate representation, public release commitment, and backbone-controlled ablations are concrete engineering contributions that lower the barrier to reproducible multi-format dissemination research. Even modest improvements in suite-level faithfulness would have practical value for researchers and conference organizers.

major comments (3)
  1. §4.2 and §5 (Tabs. 1–3, Eq. 7): Nearly all primary metrics (FactScore extraction+entailment, QA pair generation, readability, visual-content accuracy, aesthetics, layout) are produced by Gemini-3-Flash (or close family variants). The same model family is the strongest OmniPresent backbone and yields the largest reported deltas (e.g., slides FactScore 0.816 vs. 0.666/0.652). This creates a self-preference risk that is not broken by the ablations or inclusion tables, which inherit the identical judge. An independent human audit of atomic-fact coverage and cross-format numerical consistency on a non-trivial subset, or a second-family judge (e.g., GPT/Claude), is required before the faithfulness claims can be treated as established.
  2. App. E / Tab. 7: The 13-participant Likert study validates only visual dimensions and shows rank correlation with the VLM rubric; it supplies no independent check of factual fidelity, claim consistency, or QA accuracy. Because the central claim is suite-level faithfulness (not merely aesthetics), the absence of human fact-level validation leaves the largest metric gains (FactScore, QA) without external corroboration.
  3. §4.1 and App. A.1: The evaluation test set is only 50 papers (stratified from 1,037). While the SPECTER2+K=20 sampling is carefully described, 50 is small for claiming systematic superiority across venues, years, and topics; confidence intervals or bootstrap estimates are not reported. The out-of-domain set (Tab. 5) is even smaller (n=10). Expanding the test set or providing variance estimates would strengthen the generalizability claim.
minor comments (5)
  1. Fig. 1 and Tab. 1 caption: Compression Rate is listed but never formally defined in the main text; a short equation or pointer to the appendix would help.
  2. Eq. 1: The containment relation I({POSTER,PAGE}) ⊆ I(SLIDES) ⊆ I(PAPER) is presented as definitional; a brief justification or empirical check that human suites actually obey it would be useful.
  3. §3.5 / Video Gen: The video pipeline is described only briefly; how speech-script synchronization is enforced after gap-filling (modal synchronization claim) could be clarified with a short pseudocode step.
  4. References: Several concurrent systems (PaperX, EvoPresent, AutoPage) appear as arXiv preprints; ensure final versions or DOIs are updated if available at camera-ready.
  5. Typographical: “OmniPresentto” (Abstract) and occasional missing spaces around math (e.g., p f) should be cleaned.

Circularity Check

0 steps flagged

No circular derivation chain; empirical multi-agent systems paper whose claims rest on external baselines, ablations, and a held-out multi-format benchmark rather than self-definitional or fitted-as-prediction steps.

full rationale

OmniPresent is an engineering/systems contribution that formalizes unified presentation-suite generation, introduces a Plan-Verify-Render pipeline over a shared Knowledge Stream K and HTML IR, and evaluates it on OmniPreBench (1 037 paired papers + 50-paper stratified test set) against format-specialized systems (Paper2Poster, EvoPresent, AutoPage, PaperX) and matched-backbone end-to-end LLMs. The four-stage pipeline (parse → format-specific planning → cross-artifact verify-and-repair → layout/render) and the containment relation I(POSTER,PAGE) ⊆ I(SLIDES) ⊆ I(PAPER) are design choices, not mathematical derivations that reduce outputs to inputs by construction. FactScore (Eq. 7), QA accuracy, readability, visual rubrics, inclusion scores (Tab. 4) and conflict-reduction rates (Tab. 6) are computed against source-paper evidence or human-authored artifacts; the CV ablation (Tab. 3) removes a module and measures drops under identical decoding. No parameter is fitted to a target quantity and then re-presented as a prediction of that quantity; no uniqueness theorem or ansatz is imported solely via self-citation to force the result; self-citations (e.g., AutoPage) appear only as baselines or related work. The dual use of Gemini-family models as both backbone and VLM judge is an evaluation-contamination risk, not a circular reduction of the claimed method. Consequently the paper exhibits none of the six circularity patterns and receives score 0.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 3 invented entities

Load-bearing structure is engineering design choices and evaluation assumptions rather than free physical constants. The method assumes PDF parsing fidelity, that a linearized knowledge stream plus claim sets suffice for consistency, that the poster/page ⊆ slides salience rule is the right suite constraint, and that VLM judges track human notions of faithfulness and aesthetics. Free parameters are operational (iteration budget T, cluster count K=20, Likert scales). Invented entities are the Knowledge Stream and the formalized Cross-Artifact Verification loop as first-class suite mechanisms.

free parameters (3)
  • max verify-repair iterations T
    Algorithm 1 stops the critic-editor loop after T iterations or empty delta; T is a hand-chosen computational budget that can leave residual conflicts if exhausted.
  • K=20 SPECTER2 clusters for testset sampling
    Appendix A.1 sets K-Means K=20 for topic coverage when selecting the 50-paper test set; different K changes which papers enter evaluation.
  • LLM/VLM judge prompts and scoring rubrics
    FactScore, readability, aesthetics, and layout scores depend on Gemini-3-Flash (and similar) prompts and 1–5 rubrics that are design choices of the evaluation, not derived quantities.
axioms (5)
  • domain assumption Renderable HTML (DOM + CSS) is a sufficient universal intermediate representation for posters, slides, pages, and video-from-slides.
    Stated as a key design decision in §3 and Discussion; underpins centralized planning and code-level verification.
  • ad hoc to paper Salience containment I({POSTER,PAGE}) ⊆ I(SLIDES) ⊆ I(PAPER) is the correct suite-level coverage constraint.
    Eq. 1 and Gap Check (Eq. 5) encode this; it drives what the verifier forces into slides.
  • domain assumption Atomic claims extracted by an LLM from the paper, linked via evidence index E, are adequate units for conflict and gap detection.
    Stage 1–3 depend on claim extraction and entailment checks against E.
  • domain assumption VLM/LLM-as-judge metrics (FactScore, Q&A, aesthetics, layout) validly rank presentation quality and faithfulness.
    §4.2 evaluation protocol; partially supported by a small human preference study in §5.3.
  • domain assumption MinerU (and related tools) extract text/figures/tables from PDFs with enough fidelity for evidence grounding.
    Stage 1 pipeline foundation; errors here propagate to all formats.
invented entities (3)
  • Knowledge Stream K as elementary units (c, e, t) no independent evidence
    purpose: Single modality-agnostic source of truth that all format planners query to reduce cross-format drift.
    Defined in §3.2; linearized for LLM prompting in App. D. Independent evidence is only internal empirical gains, not an external physical handle.
  • Cross-Artifact Verification (critic + editor) loop with Δ_conflict and Δ_gap no independent evidence
    purpose: Detect unsupported/contradictory claims and missing poster/page content in slides, then repair plans before render.
    Core method contribution in §3.4 and Algorithm 1; ablations show metric drops when removed.
  • OmniPreBench paired multi-format suite dataset and VLM evaluation protocol independent evidence
    purpose: Enable scalable comparison of suite-level generation methods.
    §4; constructed from public conference artifacts 2023–2025 with completeness filters.

pith-pipeline@v1.1.0-grok45 · 27623 in / 3817 out tokens · 38694 ms · 2026-07-12T09:29:24.079940+00:00 · methodology

0 comments
read the original abstract

Transforming static research papers into dynamic media such as posters, slides, and videos is essential for effective dissemination but remains a labor-intensive challenge. Existing automated approaches often treat these formats in isolation and consequently fail to maintain semantic consistency across the entire presentation suite. We address this fragmentation by formalizing the task of unified presentation suite generation and proposing $\textbf{OmniPresent}$ to orchestrate the creation of coherent deliverables. Our framework adopts a renderable HTML representation to enable centralized content planning and a self-correcting verify-and-repair loop that actively resolves conflicts across modalities. We further facilitate scalable research in this domain by releasing $\textbf{OmniPreBench}$, a comprehensive dataset comprising over one thousand papers with paired artifacts, and establishing a rigorous VLM-based evaluation protocol. Empirical results confirm that our method generates high-quality and faithful presentation suites that significantly surpass strong baselines in both accuracy and visual appeal.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

71 extracted references · 33 linked inside Pith

  1. [1]

    https://mistral.ai/news/mistral-small-3-1

    Mistral Small 3.1 | Mistral AI — mistral.ai. https://mistral.ai/news/mistral-small-3-1 . [Accessed 15-10- 2025]

    2025

  2. [2]

    [Accessed 15-10-2025]

    Grok 4 | xAI — x.ai.https://x.ai/news/grok-4. [Accessed 15-10-2025]

    2025

  3. [3]

    Phi-4 technical report.arXiv preprint arXiv:2412.08905, 2024

    Marah Abdin, Jyoti Aneja, Harkirat Behl, Sébastien Bubeck, Ronen Eldan, Suriya Gunasekar, Michael Harrison, Russell J Hewett, Mojan Javaheripi, Piero Kauffmann, et al. Phi-4 technical report.arXiv preprint arXiv:2412.08905, 2024

    Pith/arXiv arXiv 2024

  4. [4]

    Gpt-4 technical report.arXiv preprint arXiv:2303.08774, 2023

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report.arXiv preprint arXiv:2303.08774, 2023

    Pith/arXiv arXiv 2023

  5. [5]

    Evaluating the evaluators: Are readability metrics good measures of readability?, 2025

    Isabel Cachola, Daniel Khashabi, and Mark Dredze. Evaluating the evaluators: Are readability metrics good measures of readability?, 2025. URLhttps://arxiv.org/abs/2508.19221

    Pith/arXiv arXiv 2025

  6. [6]

    LangChain, October 2022

    Harrison Chase. LangChain, October 2022. URLhttps://github.com/langchain-ai/langchain

    2022

  7. [7]

    Autopr: Let’s automate your academic promotion!, 2025

    Qiguang Chen, Zheng Yan, Mingda Yang, Libo Qin, Yixin Yuan, Hanjing Li, Jinhao Liu, Yiyan Ji, Dengyun Peng, Jiannan Guan, Mengkang Hu, Yantao Du, and Wanxiang Che. Autopr: Let’s automate your academic promotion!, 2025. URL https://arxiv.org/abs/2510.09558

    arXiv 2025

  8. [8]

    Chain-of- verification reduces hallucination in large language models

    Shehzaad Dhuliawala, Mojtaba Komeili, Jing Xu, Roberta Raileanu, Xian Li, Asli Celikyilmaz, and Jason Weston. Chain-of- verification reduces hallucination in large language models. InFindings of the Association for Computational Linguistics: ACL 2024, pages 3563–3578, Bangkok, Thailand, August 2024. Association for Computational Linguistics. doi: 10.18653...

    doi:10.18653/v1/2024 2024

  9. [9]

    Doc2ppt: Automatic presentation slides generation from scientific documents

    Tsu-Jui Fu, William Yang Wang, Daniel McDuff, and Yale Song. Doc2ppt: Automatic presentation slides generation from scientific documents. InProceedings of the AAAI Conference on Artificial Intelligence, volume 36, pages 634–642, 2022

    2022

  10. [10]

    Improving language model negotiation with self-play and in-context learning from ai feedback.arXiv preprint arXiv:2305.10142, 2023

    Yao Fu, Hao Peng, Tushar Khot, and Mirella Lapata. Improving language model negotiation with self-play and in-context learning from ai feedback.arXiv preprint arXiv:2305.10142, 2023

    Pith/arXiv arXiv 2023

  11. [11]

    Autopresent: Designing structured visuals from scratch

    Jiaxin Ge, Zora Zhiruo Wang, Xuhui Zhou, Yi-Hao Peng, Sanjay Subramanian, Qinyue Tan, Maarten Sap, Alane Suhr, Daniel Fried, Graham Neubig, et al. Autopresent: Designing structured visuals from scratch. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 2902–2911, 2025

    2025

  12. [12]

    Chatglm: A family of large language models from glm-130b to glm-4 all tools, 2024

    Team GLM, Aohan Zeng, Bin Xu, Bowen Wang, Chenhui Zhang, Da Yin, Diego Rojas, Guanyu Feng, Hanlin Zhao, Hanyu Lai, Hao Yu, Hongning Wang, Jiadai Sun, Jiajie Zhang, Jiale Cheng, Jiayi Gui, Jie Tang, Jing Zhang, Juanzi Li, Lei Zhao, Lindong Wu, Lucen Zhong, Mingdao Liu, Minlie Huang, Peng Zhang, Qinkai Zheng, Rui Lu, Shuaiqi Duan, Shudan Zhang, Shulin Cao, ...

    2024

  13. [13]

    Ppsgen: Learning to generate presentation slides for academic papers

    Yue Hu and Xiaojun Wan. Ppsgen: Learning to generate presentation slides for academic papers. InIJCAI, pages 2099–2105, 2013

    2099

  14. [14]

    Understanding the planning of llm agents: A survey.arXiv preprint arXiv:2402.02716, 2024

    Xu Huang, Weiwen Liu, Xiaolong Chen, Xingmei Wang, Hao Wang, Defu Lian, Yasheng Wang, Ruiming Tang, and Enhong Chen. Understanding the planning of llm agents: A survey.arXiv preprint arXiv:2402.02716, 2024

    Pith/arXiv arXiv 2024

  15. [15]

    VBench++: Comprehensive and versatile benchmark suite for video generative models.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

    Ziqi Huang, Fan Zhang, Xiaojie Xu, Yinan He, Jiashuo Yu, Ziyue Dong, Qianli Ma, Nattapol Chanpaisit, Chenyang Si, Yuming Jiang, Yaohui Wang, Xinyuan Chen, Ying-Cong Chen, Limin Wang, Dahua Lin, Yu Qiao, and Ziwei Liu. VBench++: Comprehensive and versatile benchmark suite for video generative models.IEEE Transactions on Pattern Analysis and Machine Intelli...

    doi:10.1109/tpami.2025.3633890 2025

  16. [16]

    Talk to your slides: Language-driven agents for efficient slide editing, 2025

    Kyudan Jung, Hojun Cho, Jooyeol Yun, Soyoung Yang, Jaehyeok Jang, and Jaegul Choo. Talk to your slides: Language-driven agents for efficient slide editing, 2025. URLhttps://arxiv.org/abs/2505.11604

    Pith/arXiv arXiv 2025

  17. [17]

    Slidespawn: An automatic slides generation system for research publications

    Keshav Kumar and Ravindranath Chowdary. Slidespawn: An automatic slides generation system for research publications. arXiv preprint arXiv:2411.17719, 2024

    Pith/arXiv arXiv 2024

  18. [18]

    Prometheus-vision: Vision-language model as a judge for fine-grained evaluation, 2024

    Seongyun Lee, Seungone Kim, Sue Hyun Park, Geewook Kim, and Minjoon Seo. Prometheus-vision: Vision-language model as a judge for fine-grained evaluation, 2024

    2024

  19. [19]

    Search-o1: Agentic search-enhanced large reasoning models.arXiv preprint arXiv:2501.05366, 2025

    Xiaoxi Li, Guanting Dong, Jiajie Jin, Yuyao Zhang, Yujia Zhou, Yutao Zhu, Peitian Zhang, and Zhicheng Dou. Search-o1: Agentic search-enhanced large reasoning models.arXiv preprint arXiv:2501.05366, 2025. 21

    Pith/arXiv arXiv 2025

  20. [20]

    Webthinker: Empowering large reasoning models with deep research capability.arXiv preprint arXiv:2504.21776, 2025

    Xiaoxi Li, Jiajie Jin, Guanting Dong, Hongjin Qian, Yutao Zhu, Yongkang Wu, Ji-Rong Wen, and Zhicheng Dou. Webthinker: Empowering large reasoning models with deep research capability.arXiv preprint arXiv:2504.21776, 2025

    Pith/arXiv arXiv 2025

  21. [21]

    Slidegen: Collaborative multimodal agents for scientific slide generation.arXiv preprint arXiv:2512.04529, 2025

    Xin Liang, Xiang Zhang, Yiwei Xu, Siqi Sun, and Chenyu You. Slidegen: Collaborative multimodal agents for scientific slide generation.arXiv preprint arXiv:2512.04529, 2025

    arXiv 2025

  22. [22]

    Llm-eval: Unified multi-dimensional automatic evaluation for open-domain conversations with large language models.arXiv preprint arXiv:2305.13711, 2023

    Yen-Ting Lin and Yun-Nung Chen. Llm-eval: Unified multi-dimensional automatic evaluation for open-domain conversations with large language models.arXiv preprint arXiv:2305.13711, 2023

    Pith/arXiv arXiv 2023

  23. [23]

    Presenting a paper is an art: Self-improvement aesthetic agents for academic presentations, 2025

    Chengzhi Liu, Yuzhe Yang, Kaiwen Zhou, Zhen Zhang, Yue Fan, Yanan Xie, Peng Qi, and Xin Eric Wang. Presenting a paper is an art: Self-improvement aesthetic agents for academic presentations, 2025. URL https://arxiv.org/abs/2510. 05571

    2025

  24. [24]

    Mitigating hallucination in large multi-modal models via robust instruction tuning.arXiv preprint arXiv:2306.14565, 2023

    Fuxiao Liu, Kevin Lin, Linjie Li, Jianfeng Wang, Yaser Yacoob, and Lijuan Wang. Mitigating hallucination in large multi-modal models via robust instruction tuning.arXiv preprint arXiv:2306.14565, 2023

    Pith/arXiv arXiv 2023

  25. [25]

    Visual instruction tuning.Advances in neural information processing systems, 36:34892–34916, 2023

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning.Advances in neural information processing systems, 36:34892–34916, 2023

    2023

  26. [27]

    Preacher: Paper-to-video agentic system, 2025

    Jingwei Liu, Ling Yang, Hao Luo, Fan Wang, Hongyan Li, and Mengdi Wang. Preacher: Paper-to-video agentic system, 2025. URLhttps://arxiv.org/abs/2508.09632

    Pith/arXiv arXiv 2025

  27. [28]

    Human-agent collaborative paper-to-page crafting for under $0.1, 2025

    Qianli Ma, Siyu Wang, Yilin Chen, Yinhao Tang, Yixiang Yang, Chang Guo, Bingjie Gao, Zhening Xing, Yanan Sun, and Zhipeng Zhang. Human-agent collaborative paper-to-page crafting for under $0.1, 2025. URL https://arxiv.org/abs/ 2510.19600

    Pith/arXiv arXiv 2025

  28. [29]

    Paper2rebuttal: A multi-agent framework for transparent author response assistance, 2026

    Qianli Ma, Chang Guo, Zhiheng Tian, Siyu Wang, Jipeng Xiao, Yuanhao Yue, and Zhipeng Zhang. Paper2rebuttal: A multi-agent framework for transparent author response assistance, 2026. URL https://arxiv.org/abs/2601.14171

    Pith/arXiv arXiv 2026

  29. [30]

    Self-refine: Iterative refinement with self-feedback, 2023

    Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, Shashank Gupta, Bodhisattwa Prasad Majumder, Katherine Hermann, Sean Welleck, Amir Yazdanbakhsh, and Peter Clark. Self-refine: Iterative refinement with self-feedback, 2023. URL https://arxiv.org/ abs/2303.17651

    Pith/arXiv arXiv 2023

  30. [31]

    The llama 3.1 herd of models, 2024.https://huggingface.co/meta-llama

    Meta AI. The llama 3.1 herd of models, 2024.https://huggingface.co/meta-llama

    2024

  31. [32]

    Davis, Yaohui Zhang, Jonathan K

    Jiacheng Miao, Joe R. Davis, Yaohui Zhang, Jonathan K. Pritchard, and James Zou. Paper2agent: Reimagining research papers as interactive and reliable ai agents, 2025. URLhttps://arxiv.org/abs/2509.06917

    arXiv 2025

  32. [33]

    Factscore: Fine-grained atomic evaluation of factual precision in long form text generation

    Sewon Min, Kalpesh Krishna, Xinxi Lyu, Mike Lewis, Wen-tau Yih, Pang Koh, Mohit Iyyer, Luke Zettlemoyer, and Hannaneh Hajishirzi. Factscore: Fine-grained atomic evaluation of factual precision in long form text generation. InProceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 12076–12100, 2023

    2023

  33. [34]

    Gpt-5 system card

    OpenAI. Gpt-5 system card. Technical report, August 2025. URL https://openai.com/index/ gpt-5-system-card. Available at: https://cdn.openai.com/gpt-5-system-card.pdf

    2025

  34. [35]

    gpt-oss-120b & gpt-oss-20b model card, 2025

    OpenAI. gpt-oss-120b & gpt-oss-20b model card, 2025. URLhttps://arxiv.org/abs/2508.10925

    Pith/arXiv arXiv 2025

  35. [36]

    Paper2poster: Towards multimodal poster automation from scientific papers.arXiv preprint arXiv:2505.21497, 2025

    Wei Pang, Kevin Qinghong Lin, Xiangru Jian, Xi He, and Philip Torr. Paper2poster: Towards multimodal poster automation from scientific papers.arXiv preprint arXiv:2505.21497, 2025

    arXiv 2025

  36. [37]

    Tailored summary for automatic poster generator

    Kanya Paramita and Masayu Leylia Khodra. Tailored summary for automatic poster generator. In2016 International Conference On Advanced Informatics: Concepts, Theory And Application (ICAICTA), pages 1–6. IEEE, 2016

    2016

  37. [38]

    Tool learning with large language models: A survey.Frontiers of Computer Science, 19(8):198343, 2025

    Changle Qu, Sunhao Dai, Xiaochi Wei, Hengyi Cai, Shuaiqiang Wang, Dawei Yin, Jun Xu, and Ji-Rong Wen. Tool learning with large language models: A survey.Frontiers of Computer Science, 19(8):198343, 2025

    2025

  38. [39]

    Paper2code: Automating code generation from scientific papers in machine learning.arXiv preprint arXiv:2504.17192, 2025

    Minju Seo, Jinheon Baek, Seongyun Lee, and Sung Ju Hwang. Paper2code: Automating code generation from scientific papers in machine learning.arXiv preprint arXiv:2504.17192, 2025

    arXiv 2025

  39. [40]

    Presentagent: Multimodal agent for presentation video generation.arXiv preprint arXiv:2507.04036, 2025

    Jingwei Shi, Zeyu Zhang, Biao Wu, Yanjie Liang, Meng Fang, Ling Chen, and Yang Zhao. Presentagent: Multimodal agent for presentation video generation.arXiv preprint arXiv:2507.04036, 2025. 22

    Pith/arXiv arXiv 2025

  40. [41]

    Tool learning in the wild: Empowering language models as automatic tool agents

    Zhengliang Shi, Shen Gao, Lingyong Yan, Yue Feng, Xiuyi Chen, Zhumin Chen, Dawei Yin, Suzan Verberne, and Zhaochun Ren. Tool learning in the wild: Empowering language models as automatic tool agents. InProceedings of the ACM on Web Conference 2025, pages 2222–2237, 2025

    2025

  41. [42]

    Reflexion: Language agents with verbal reinforcement learning.Advances in Neural Information Processing Systems, 36:8634–8652, 2023

    Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. Reflexion: Language agents with verbal reinforcement learning.Advances in Neural Information Processing Systems, 36:8634–8652, 2023

    2023

  42. [43]

    Design2Code: Benchmarking mul- timodal code generation for automated front-end engineering

    Chenglei Si, Yanzhe Zhang, Ryan Li, Zhengyuan Yang, Ruibo Liu, and Diyi Yang. Design2Code: Benchmarking mul- timodal code generation for automated front-end engineering. In Luis Chiruzzo, Alan Ritter, and Lu Wang, editors, Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Lin- guistics: Human La...

    2025

  43. [44]

    ISBN 979-8-89176-189-6

    Association for Computational Linguistics. ISBN 979-8-89176-189-6. doi: 10.18653/v1/2025.naacl-long.199. URL https://aclanthology.org/2025.naacl-long.199/

    doi:10.18653/v1/2025.naacl-long.199 2025

  44. [45]

    D2s: Document-to-slide generation via query-based text summarization.arXiv preprint arXiv:2105.03664, 2021

    Edward Sun, Yufang Hou, Dakuo Wang, Yunfeng Zhang, and Nancy XR Wang. D2s: Document-to-slide generation via query-based text summarization.arXiv preprint arXiv:2105.03664, 2021

    Pith/arXiv arXiv 2021

  45. [46]

    Adaplanner: Adaptive planning from feedback with language models.Advances in neural information processing systems, 36:58202–58245, 2023

    Haotian Sun, Yuchen Zhuang, Lingkai Kong, Bo Dai, and Chao Zhang. Adaplanner: Adaptive planning from feedback with language models.Advances in neural information processing systems, 36:58202–58245, 2023

    2023

  46. [47]

    P2p: Automated paper-to-poster generation and fine-grained benchmark.arXiv preprint arXiv:2505.17104, 2025

    Tao Sun, Enhao Pan, Zhengkai Yang, Kaixin Sui, Jiajun Shi, Xianfu Cheng, Tongliang Li, Wenhao Huang, Ge Zhang, Jian Yang, et al. P2p: Automated paper-to-poster generation and fine-grained benchmark.arXiv preprint arXiv:2505.17104, 2025

    Pith/arXiv arXiv 2025

  47. [48]

    Docling technical report

    Deep Search Team. Docling technical report. Technical report, 8 2024. URL https://arxiv.org/abs/2408.09869

    Pith/arXiv arXiv 2024

  48. [49]

    Gemini: a family of highly capable multimodal models.arXiv preprint arXiv:2312.11805, 2023

    Gemini Team, Rohan Anil, Sebastian Borgeaud, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, Katie Millican, et al. Gemini: a family of highly capable multimodal models.arXiv preprint arXiv:2312.11805, 2023

    Pith/arXiv arXiv 2023

  49. [50]

    Multi-agent collaboration mechanisms: A survey of llms.arXiv preprint arXiv:2501.06322, 2025

    Khanh-Tung Tran, Dung Dao, Minh-Duong Nguyen, Quoc-Viet Pham, Barry O’Sullivan, and Hoang D Nguyen. Multi-agent collaboration mechanisms: A survey of llms.arXiv preprint arXiv:2501.06322, 2025

    Pith/arXiv arXiv 2025

  50. [51]

    Mineru: An open-source solution for precise document content extraction, 2024

    Bin Wang, Chao Xu, Xiaomeng Zhao, Linke Ouyang, Fan Wu, Zhiyuan Zhao, Rui Xu, Kaiwen Liu, Yuan Qu, Fukai Shang, Bo Zhang, Liqun Wei, Zhihao Sui, Wei Li, Botian Shi, Yu Qiao, Dahua Lin, and Conghui He. Mineru: An open-source solution for precise document content extraction, 2024. URLhttps://arxiv.org/abs/2409.18839

    Pith/arXiv arXiv 2024

  51. [52]

    V oyager: An open-ended embodied agent with large language models.arXiv preprint arXiv:2305.16291, 2023

    Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and Anima Anandkumar. V oyager: An open-ended embodied agent with large language models.arXiv preprint arXiv:2305.16291, 2023

    Pith/arXiv arXiv 2023

  52. [53]

    A survey on large language model based autonomous agents.Frontiers of Computer Science, 18(6):186345, 2024

    Lei Wang, Chen Ma, Xueyang Feng, Zeyu Zhang, Hao Yang, Jingsen Zhang, Zhiyuan Chen, Jiakai Tang, Xu Chen, Yankai Lin, et al. A survey on large language model based autonomous agents.Frontiers of Computer Science, 18(6):186345, 2024

    2024

  53. [54]

    Autosurvey: Large language models can automatically write surveys.Advances in neural information processing systems, 37:115119–115145, 2024

    Yidong Wang, Qi Guo, Wenjin Yao, Hongbo Zhang, Xin Zhang, Zhen Wu, Meishan Zhang, Xinyu Dai, Qingsong Wen, Wei Ye, et al. Autosurvey: Large language models can automatically write surveys.Advances in neural information processing systems, 37:115119–115145, 2024

    2024

  54. [55]

    Cycleresearcher: Improving automated research via automated review

    Yixuan Weng, Minjun Zhu, Guangsheng Bao, Hongbo Zhang, Jindong Wang, Yue Zhang, and Linyi Yang. Cycleresearcher: Improving automated research via automated review. InThe Thirteenth International Conference on Learning Representations,

  55. [56]

    URLhttps://openreview.net/forum?id=bjcsVLoHYs

  56. [57]

    Autogen: Enabling next-gen llm applications via multi-agent conversations

    Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Beibin Li, Erkang Zhu, Li Jiang, Xiaoyun Zhang, Shaokun Zhang, Jiale Liu, et al. Autogen: Enabling next-gen llm applications via multi-agent conversations. InFirst Conference on Language Modeling, 2024

    2024

  57. [58]

    The rise and potential of large language model based agents: A survey.Science China Information Sciences, 68(2): 121101, 2025

    Zhiheng Xi, Wenxiang Chen, Xin Guo, Wei He, Yiwen Ding, Boyang Hong, Ming Zhang, Junzhe Wang, Senjie Jin, Enyu Zhou, et al. The rise and potential of large language model based agents: A survey.Science China Information Sciences, 68(2): 121101, 2025

    2025

  58. [59]

    Large multimodal agents: A survey.arXiv preprint arXiv:2402.15116, 2024

    Junlin Xie, Zhihong Chen, Ruifei Zhang, Xiang Wan, and Guanbin Li. Large multimodal agents: A survey.arXiv preprint arXiv:2402.15116, 2024

    Pith/arXiv arXiv 2024

  59. [61]

    Posterbot: A system for generating posters of scientific papers with neural models

    Sheng Xu and Xiaojun Wan. Posterbot: A system for generating posters of scientific papers with neural models. 2022

    2022

  60. [62]

    Pregenie: An agentic framework for high-quality visual presentation generation, 2025

    Xiaojie Xu, Xinli Xu, Sirui Chen, Haoyu Chen, Fan Zhang, and Ying-Cong Chen. Pregenie: An agentic framework for high-quality visual presentation generation, 2025. URLhttps://arxiv.org/abs/2505.21660. 23

    Pith/arXiv arXiv 2025

  61. [63]

    Textual-to-visual iterative self-verification for slide generation, 2025

    Yunqing Xu, Xinbei Ma, Jiyang Qiu, and Hai Zhao. Textual-to-visual iterative self-verification for slide generation, 2025. URL https://arxiv.org/abs/2502.15412

    Pith/arXiv arXiv 2025

  62. [64]

    Qwen3 technical report.arXiv preprint arXiv:2505.09388, 2025

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jing Zhou, Jingren Zhou, Junyang Lin, Kai Dang, Keqin Bao, Kexin Yang, ...

    Pith/arXiv arXiv 2025

  63. [65]

    React: Synergizing reasoning and acting in language models

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. React: Synergizing reasoning and acting in language models. InInternational Conference on Learning Representations (ICLR), 2023

    2023

  64. [66]

    Paperx: A unified framework for multimodal academic presentation generation with scholar dag, 2026

    Tao Yu, Minghui Zhang, Zhiqing Cui, Hao Wang, Zhongtian Luo, Shenghua Chai, Junhao Gong, Yuzhao Peng, Yuxuan Zhou, Yujia Yang, Zhenghao Zhang, Haopeng Jin, Xinming Wang, Yufei Xiong, Jiabing Yang, Jiahao Yuan, Hanqing Wang, Hongzhu Yi, Yan Huang, and Liang Wang. Paperx: A unified framework for multimodal academic presentation generation with scholar dag, ...

    arXiv 2026

  65. [67]

    Simple fast algorithms for the editing distance between trees and related problems.SIAM J

    Kaizhong Zhang and Dennis Shasha. Simple fast algorithms for the editing distance between trees and related problems.SIAM J. Comput., 18:1245–1262, 1989. URLhttps://api.semanticscholar.org/CorpusID:10970317

    1989

  66. [68]

    Postergen: Aesthetic-aware paper-to-poster generation via multi-agent llms, 2025

    Zhilin Zhang, Xiang Zhang, Jiaqi Wei, Yiwei Xu, and Chenyu You. Postergen: Aesthetic-aware paper-to-poster generation via multi-agent llms, 2025. URLhttps://arxiv.org/abs/2508.17188

    Pith/arXiv arXiv 2025

  67. [69]

    A survey of large language models.arXiv preprint arXiv:2303.18223, 1(2), 2023

    Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, Xiaolei Wang, Yupeng Hou, Yingqian Min, Beichen Zhang, Junjie Zhang, Zican Dong, et al. A survey of large language models.arXiv preprint arXiv:2303.18223, 1(2), 2023

    Pith/arXiv arXiv 2023

  68. [70]

    Pptagent: Generating and evaluating presentations beyond text-to-slides.arXiv preprint arXiv:2501.03936, 2025

    Hao Zheng, Xinyan Guan, Hao Kong, Jia Zheng, Weixiang Zhou, Hongyu Lin, Yaojie Lu, Ben He, Xianpei Han, and Le Sun. Pptagent: Generating and evaluating presentations beyond text-to-slides.arXiv preprint arXiv:2501.03936, 2025

    Pith/arXiv arXiv 2025

  69. [71]

    Judgelm: Fine-tuned large language models are scalable judges.arXiv preprint arXiv:2310.17631, 2023

    Lianghui Zhu, Xinggang Wang, and Xinlong Wang. Judgelm: Fine-tuned large language models are scalable judges.arXiv preprint arXiv:2310.17631, 2023

    Pith/arXiv arXiv 2023

  70. [72]

    Deepreview: Improving llm-based paper review with human-like deep thinking process, 2025

    Minjun Zhu, Yixuan Weng, Linyi Yang, and Yue Zhang. Deepreview: Improving llm-based paper review with human-like deep thinking process, 2025. URLhttps://arxiv.org/abs/2503.08569

    Pith/arXiv arXiv 2025

  71. [74]

    Paper2video: Automatic video generation from scientific papers.arXiv preprint arXiv:2510.05096, 2025

    Zeyu Zhu, Kevin Qinghong Lin, and Mike Zheng Shou. Paper2video: Automatic video generation from scientific papers.arXiv preprint arXiv:2510.05096, 2025. 24

    arXiv 2025