pith. sign in

arxiv: 2606.17162 · v1 · pith:E37JHG7Znew · submitted 2026-06-15 · 💻 cs.CL · cs.HC· cs.MA

MemSlides: A Hierarchical Memory Driven Agent Framework for Personalized Slide Generation with Multi-turn Local Revision

Ye Jin , Yangyang Xu , Jun Zhu , Yibo Yang This is my paper

Pith reviewed 2026-06-27 03:39 UTC · model grok-4.3

classification 💻 cs.CL cs.HCcs.MA
keywords personalized slide generationhierarchical memoryagent frameworkmulti-turn revisionuser profile memoryworking memorytool memorylocal editing
0
0 comments X

The pith

Separating persistent user profiles, session working memory, and reusable tool experience enables stable personalization and reliable local edits in slide generation.

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

The paper introduces a memory organization for agents that generate and revise presentations over multiple turns. Long-term memory splits into user profile memory that stores intent-based preferences for initial output and tool memory that holds reusable editing steps, while working memory tracks active constraints within a session. This design is combined with revisions that target only the affected slide region instead of regenerating the full deck. Controlled tests show gains in matching user personas and in successful closed-loop modifications, with cases confirming preference carryover. A reader cares because most current agents lose consistency when preferences accumulate or when edits must stay precise across repeated interactions.

Core claim

MemSlides separates long-term memory into user profile memory for round-0 personalization and tool memory for execution experience, keeps these distinct from session working memory, and pairs the hierarchy with scoped slide-local revision so that updates affect only the smallest changed region; experiments indicate this yields better persona alignment on multi-persona banks, improved closed-loop modify reliability, and preference carryover across rounds.

What carries the argument

Hierarchical memory framework that divides long-term memory into intent-conditioned user profile memory and reusable tool memory while isolating session working memory, used together with scoped local revision.

If this is right

  • User profile memory produces higher persona-alignment judgments across multi-persona, multi-intent profile banks.
  • Tool-memory injection raises closed-loop modify success rates in matched-pair diagnostic settings.
  • Working memory preserves newly introduced preferences and constraints from one revision round to the next.
  • Scoped slide-local revision confines changes to the smallest affected region rather than requiring full-deck regeneration.

Where Pith is reading between the lines

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

  • The same three-way memory split may reduce error accumulation in other multi-turn agent tasks that require both stable user intent and precise iterative edits.
  • Without explicit separation, session constraints could overwrite or dilute long-term profile information in extended interactions.
  • Real-user longitudinal tests over dozens of turns would reveal whether independence between the memory stores holds under natural preference drift.

Load-bearing premise

The three memory components can be maintained and accessed independently without interference or information loss across multiple revision turns.

What would settle it

A matched experiment in which a single unified memory baseline produces equivalent or better persona-alignment scores and closed-loop modification success rates than the separated-memory version on the same profile bank and diagnostic edit pairs.

Figures

Figures reproduced from arXiv: 2606.17162 by Jun Zhu, Yangyang Xu, Ye Jin, Yibo Yang.

Figure 1
Figure 1. Figure 1: Overview of MemSlides. The framework comprises long-term memory and working [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Localized modify execution in MemSlides. Working memory supplies active temporary [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: User profile memory lifecycle in MemSlides. Long term memory stores intent-conditioned [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Tool memory flow in MemSlides. Round-scope task experience is available at job start, [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative illustration of localized modify execution. Given the same local edit request, [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative cross-job profile consolidation. Across six repeated jobs, local feedback cues [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Template-guided generation examples: selected template slides (top) and matched per [PITH_FULL_IMAGE:figures/full_fig_p021_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Illustrative qualitative trajectory for localized modify behavior. The example contrasts [PITH_FULL_IMAGE:figures/full_fig_p022_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Working-memory cases for delayed preference carryover. A future-facing rule is stated in [PITH_FULL_IMAGE:figures/full_fig_p023_9.png] view at source ↗
read the original abstract

Personalized presentation generation requires more than conditioning on a current prompt or template: agents must preserve stable user preferences across tasks, retain newly introduced preferences and constraints during multi-turn revision, and carry out local edits reliably. We propose MemSlides, a hierarchical memory framework for personalized presentation agents that separates long-term memory from working memory and further divides long-term memory into user profile memory and tool memory. User profile memory stores intent-conditioned profiles for round-0 personalization, working memory carries active preferences and session constraints across revision rounds, and tool memory stores reusable execution experience for reliable localized editing. MemSlides pairs this memory design with scoped slide-local revision, so targeted updates act on the smallest affected region instead of repeatedly regenerating the full deck. In controlled experiments, user profile memory improves persona-alignment judgments on a multi-persona, multi-intent profile bank, tool-memory injection improves closed-loop modify behavior in diagnostic matched-pair settings, and qualitative cases illustrate working memory's ability to carryover preferences. Taken together, these results suggest that effective personalization in presentation authoring depends on separating persistent user profiles, session-level working memory, and reusable execution experience across generation and localized revision.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The manuscript proposes MemSlides, a hierarchical memory framework for personalized presentation generation agents. It separates long-term memory into user profile memory (for intent-conditioned profiles) and tool memory (for reusable execution experience), with an additional working memory component for active preferences and session constraints across multi-turn revisions. The framework is paired with scoped slide-local revision to enable targeted edits rather than full regeneration. Controlled experiments are described on a multi-persona, multi-intent profile bank and matched-pair settings, with claims that user profile memory improves persona-alignment judgments, tool memory improves closed-loop modify behavior, and working memory supports preference carryover; the results are taken to indicate that separating these memory types is key to effective personalization.

Significance. If the empirical results hold and the claimed independence of the three memory stores can be verified, the work would provide a concrete architectural pattern for maintaining stable user preferences alongside session-specific and tool-based information in agentic creative tasks. This could inform designs for multi-turn personalization beyond slide generation, particularly where persistent profiles must coexist with dynamic constraints without interference.

major comments (2)
  1. [Abstract] Abstract: the abstract asserts that 'user profile memory improves persona-alignment judgments on a multi-persona, multi-intent profile bank' and 'tool-memory injection improves closed-loop modify behavior in diagnostic matched-pair settings' yet supplies no methods, baselines, quantitative results, or statistical details. Without these, it is impossible to evaluate whether the data support the central claim that separating the three memory components produces the reported gains.
  2. [Framework description (no section or equation cited)] The manuscript provides no mechanism, pseudocode, update rules, or diagnostic test to ensure that user profile memory, working memory, and tool memory can be maintained and retrieved independently without interference or leakage across revision turns (e.g., a session constraint overwriting a persistent profile or tool experience altering active constraints). This independence is load-bearing for the claim that the hierarchical design outperforms a single shared memory; if interference occurs, the reported improvements in the multi-persona bank and matched-pair settings cannot be attributed to the separation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. The comments identify opportunities to strengthen the presentation of experimental details and the explicitness of the memory separation mechanisms. We respond to each major comment below and indicate planned revisions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the abstract asserts that 'user profile memory improves persona-alignment judgments on a multi-persona, multi-intent profile bank' and 'tool-memory injection improves closed-loop modify behavior in diagnostic matched-pair settings' yet supplies no methods, baselines, quantitative results, or statistical details. Without these, it is impossible to evaluate whether the data support the central claim that separating the three memory components produces the reported gains.

    Authors: Abstracts are concise summaries; the full methods (profile bank construction, matched-pair diagnostics), baselines (flat-memory ablations), quantitative results (alignment scores, modify success rates), and statistical details appear in the Experiments section. To improve standalone readability we will add one sentence to the abstract noting the controlled multi-persona setting and the primary metrics used to support the separation claim. revision: yes

  2. Referee: [Framework description (no section or equation cited)] The manuscript provides no mechanism, pseudocode, update rules, or diagnostic test to ensure that user profile memory, working memory, and tool memory can be maintained and retrieved independently without interference or leakage across revision turns (e.g., a session constraint overwriting a persistent profile or tool experience altering active constraints). This independence is load-bearing for the claim that the hierarchical design outperforms a single shared memory; if interference occurs, the reported improvements in the multi-persona bank and matched-pair settings cannot be attributed to the separation.

    Authors: Independence is maintained by construction: user-profile memory is written once from initial intent and remains read-only thereafter; working memory is scoped to the current session and explicitly cleared or overwritten at turn boundaries; tool memory is accessed solely via execution traces and never writes back to the other stores. Retrieval is type-scoped inside the agent prompt templates. We nevertheless agree that explicit documentation is needed and will insert pseudocode for the three update/retrieval rules plus a short diagnostic experiment measuring cross-store leakage in the revised Framework section. revision: yes

Circularity Check

0 steps flagged

No significant circularity; framework proposal evaluated via experiments without mathematical derivations

full rationale

The manuscript describes a hierarchical memory architecture (user profile memory, working memory, tool memory) paired with scoped local revision and reports empirical gains from each component in controlled experiments. No equations, first-principles derivations, or parameter-fitting steps appear in the provided text. Claims rest on experimental outcomes rather than any reduction of a 'prediction' to its own inputs by construction, self-citation chains, or renamed ansatzes. The independence assumption is an unverified design premise but is not presented as a derived result, so the circularity patterns do not apply.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 3 invented entities

The central claim rests on the conceptual separation of three memory types and the effectiveness of scoped local revision; these are introduced without independent evidence beyond the described experiments.

axioms (1)
  • domain assumption Separating long-term memory into user profile memory and tool memory plus working memory enables stable personalization and reliable local edits across turns.
    This separation is presented as the key design choice whose benefits are demonstrated in the experiments.
invented entities (3)
  • user profile memory no independent evidence
    purpose: Stores intent-conditioned profiles for initial personalization
    Newly defined component in the hierarchical memory design.
  • working memory no independent evidence
    purpose: Carries active preferences and session constraints across revision rounds
    Newly defined component in the hierarchical memory design.
  • tool memory no independent evidence
    purpose: Stores reusable execution experience for localized editing
    Newly defined component in the hierarchical memory design.

pith-pipeline@v0.9.1-grok · 5743 in / 1418 out tokens · 58615 ms · 2026-06-27T03:39:53.418821+00:00 · methodology

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

66 extracted references · 25 canonical work pages · 8 internal anchors

  1. [1]

    Do as i can, not as i say: Grounding language in robotic affordances.arXiv preprint arXiv:2204.01691, 2022

    Michael Ahn, Anthony Brohan, Noah Brown, Yevgen Chebotar, Omar Cortes, Byron David, Chelsea Finn, Chuyuan Fu, Keerthana Gopalakrishnan, Karol Hausman, et al. Do as i can, not as i say: Grounding language in robotic affordances.arXiv preprint arXiv:2204.01691, 2022. URL https://arxiv.org/ abs/2204.01691

    Pith/arXiv arXiv 2022

  2. [2]

    En- hancing presentation slide generation by LLMs with a multi-staged end-to-end approach

    Sambaran Bandyopadhyay, Himanshu Maheshwari, Anandhavelu Natarajan, and Apoorv Saxena. En- hancing presentation slide generation by LLMs with a multi-staged end-to-end approach. InProceed- ings of the 17th International Natural Language Generation Conference, pages 222–229, Tokyo, Japan, September 2024. Association for Computational Linguistics. doi: 10.1...

    work page doi:10.18653/v1/2024.inlg-main.18 2024

  3. [3]

    Improving language models by retrieving from trillions of tokens.arXiv preprint arXiv:2112.04426, 2021

    Sebastian Borgeaud, Arthur Mensch, Jordan Hoffmann, Trevor Cai, Eliza Rutherford, Katie Millican, George van den Driessche, Jean-Baptiste Lespiau, Bogdan Damoc, Aidan Clark, et al. Improving language models by retrieving from trillions of tokens.arXiv preprint arXiv:2112.04426, 2021. URL https://arxiv.org/abs/2112.04426

    Pith/arXiv arXiv 2021

  4. [4]

    Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory

    Prateek Chhikara, Dev Khant, Saket Aryan, Taranjeet Singh, and Deshraj Yadav. Mem0: Building production-ready AI agents with scalable long-term memory.arXiv preprint arXiv:2504.19413, 2025. doi: 10.48550/arXiv.2504.19413. URLhttps://arxiv.org/abs/2504.19413

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2504.19413 2025

  5. [5]

    The vendi score: A diversity evaluation metric for machine learning.Transactions on Machine Learning Research, 2023

    Dan Friedman and Adji Bousso Dieng. The vendi score: A diversity evaluation metric for machine learning.Transactions on Machine Learning Research, 2023. URL https://openreview.net/forum? id=g97OHbQyk1

    2023

  6. [6]

    CoT-VLA: Visual chain-of-thought reasoning for vision- language-action models,

    Jiaxin Ge, Zora Zhiruo Wang, Xuhui Zhou, Yi-Hao Peng, Sanjay Subramanian, Qinyue Tan, Maarten Sap, Alane Suhr, Daniel Fried, Graham Neubig, and Trevor Darrell. AutoPresent: Designing structured visuals from scratch. InProceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 2902–2911, June 2025. doi: 10.1109/CVPR52734.202...

    work page doi:10.1109/cvpr52734.2025 2025

  7. [7]

    REALM: Retrieval- augmented language model pre-training

    Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, and Ming-Wei Chang. REALM: Retrieval- augmented language model pre-training. InProceedings of the 37th International Conference on Machine Learning, 2020. URLhttps://arxiv.org/abs/2002.08909

    Pith/arXiv arXiv 2020

  8. [8]

    Agentic tool use in large language models.arXiv preprint arXiv:2604.00835, 2026

    Jinchao Hu, Meizhi Zhong, Kehai Chen, Xuefeng Bai, and Min Zhang. Agentic tool use in large language models.arXiv preprint arXiv:2604.00835, 2026. URLhttps://arxiv.org/abs/2604.00835

    arXiv 2026

  9. [9]

    LayoutDM: Discrete diffusion model for controllable layout generation.arXiv preprint arXiv:2303.08137, 2023

    Naoto Inoue, Kotaro Kikuchi, Edgar Simo-Serra, Mayu Otani, and Kota Yamaguchi. LayoutDM: Discrete diffusion model for controllable layout generation.arXiv preprint arXiv:2303.08137, 2023. URL https: //arxiv.org/abs/2303.08137

    arXiv 2023

  10. [10]

    Atlas: Few-shot learning with retrieval augmented language models.arXiv preprint arXiv:2208.03299, 2022

    Gautier Izacard, Patrick Lewis, Maria Lomeli, Lucas Hosseini, Fabio Petroni, Timo Schick, Jane Dwivedi- Yu, Armand Joulin, Sebastian Riedel, and Edouard Grave. Atlas: Few-shot learning with retrieval augmented language models.arXiv preprint arXiv:2208.03299, 2022. URL https://arxiv.org/abs/ 2208.03299

    Pith/arXiv arXiv 2022

  11. [11]

    Know me, respond to me: Benchmarking LLMs for dynamic user profiling and personalized responses at scale

    Bowen Jiang, Zhuoqun Hao, Young-Min Cho, Bryan Li, Yuan Yuan, Sihao Chen, Lyle Ungar, Camillo J. Taylor, and Dan Roth. Know me, respond to me: Benchmarking LLMs for dynamic user profiling and personalized responses at scale.arXiv preprint arXiv:2504.14225, 2025. doi: 10.48550/arXiv.2504.14225. URLhttps://arxiv.org/abs/2504.14225

    work page doi:10.48550/arxiv.2504.14225 2025

  12. [12]

    Memory OS of AI agent

    Jiazheng Kang, Mingming Ji, Zhe Zhao, and Ting Bai. Memory OS of AI agent. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 25961–25970, Suzhou, China, November 2025. Association for Computational Linguistics. doi: 10.18653/v1/2025.emnlp-main.1318. URLhttps://aclanthology.org/2025.emnlp-main.1318/. 10

    work page doi:10.18653/v1/2025.emnlp-main.1318 2025

  13. [13]

    MRKL systems: A modular, neuro-symbolic architecture that combines large language models, external knowledge sources and discrete reasoning

    Ehud Karpas, Omri Abend, Yonatan Belinkov, Barak Lenz, Opher Lieber, Nir Ratner, Yoav Shoham, Hofit Bata, Yoav Levine, Kevin Leyton-Brown, Dor Muhlgay, Noam Rozen, Erez Schwartz, Gal Shachaf, Shai Shalev-Shwartz, Amnon Shashua, and Moshe Tenenholtz. MRKL systems: A modular, neuro-symbolic architecture that combines large language models, external knowledg...

    Pith/arXiv arXiv 2022

  14. [14]

    Generalization through memorization: Nearest neighbor language models.arXiv preprint arXiv:1911.00172, 2019

    Urvashi Khandelwal, Omer Levy, Dan Jurafsky, Luke Zettlemoyer, and Mike Lewis. Generalization through memorization: Nearest neighbor language models.arXiv preprint arXiv:1911.00172, 2019. URL https://arxiv.org/abs/1911.00172

    arXiv 1911

  15. [15]

    Prometheus: Inducing fine-grained evaluation capability in language models.arXiv preprint arXiv:2310.08491, 2023

    Seungone Kim, Jamin Shin, Yejin Cho, Joel Jang, Shayne Longpre, Hwaran Lee, Sangdoo Yun, Seongjin Shin, Sungdong Kim, James Thorne, and Minjoon Seo. Prometheus: Inducing fine-grained evaluation capability in language models.arXiv preprint arXiv:2310.08491, 2023. URL https://arxiv.org/abs/ 2310.08491

    arXiv 2023

  16. [16]

    Retrieval- augmented generation for knowledge-intensive NLP tasks

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. Retrieval- augmented generation for knowledge-intensive NLP tasks. InAdvances in Neural Information Processing Systems, volume 33, pages 9459–9474, 2020. URLhttps://pr...

    2020

  17. [17]

    A persona- based neural conversation model

    Jiwei Li, Michel Galley, Chris Brockett, Georgios Spithourakis, Jianfeng Gao, and Bill Dolan. A persona- based neural conversation model. InProceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 994–1003, Berlin, Germany, August 2016. Association for Computational Linguistics. doi: 10.18653/v...

    work page doi:10.18653/v1/p16-1094 2016

  18. [18]

    API-bank: A comprehensive benchmark for tool-augmented LLMs.arXiv preprint arXiv:2304.08244, 2023

    Minghao Li, Yingxiu Zhao, Bowen Yu, Feifan Song, Hangyu Li, Haiyang Yu, Zhoujun Li, Fei Huang, and Yongbin Li. API-bank: A comprehensive benchmark for tool-augmented LLMs.arXiv preprint arXiv:2304.08244, 2023. URLhttps://arxiv.org/abs/2304.08244

    Pith/arXiv arXiv 2023

  19. [19]

    Gonzalez, and Ion Stoica

    Tianle Li, Wei-Lin Chiang, Evan Frick, Lisa Dunlap, Tianhao Wu, Banghua Zhu, Joseph E. Gonzalez, and Ion Stoica. From crowdsourced data to high-quality benchmarks: Arena-hard and benchbuilder pipeline. arXiv preprint arXiv:2406.11939, 2024. URLhttps://arxiv.org/abs/2406.11939

    Pith/arXiv arXiv 2024

  20. [20]

    A survey of personalization: From RAG to agent.arXiv preprint arXiv:2504.10147, 2025

    Xiaopeng Li, Pengyue Jia, Derong Xu, Yi Wen, Yingyi Zhang, Wenlin Zhang, Wanyu Wang, Yichao Wang, Zhaocheng Du, Xiangyang Li, Yong Liu, Huifeng Guo, Ruiming Tang, and Xiangyu Zhao. A survey of personalization: From RAG to agent.arXiv preprint arXiv:2504.10147, 2025. URL https: //arxiv.org/abs/2504.10147

    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. doi: 10.48550/arXiv.2512.04529. URLhttps://arxiv.org/abs/2512.04529

    work page doi:10.48550/arxiv.2512.04529 2025

  22. [22]

    Layout- Prompter: Awaken the design ability of large language models.arXiv preprint arXiv:2311.06495, 2023

    Jiawei Lin, Jiaqi Guo, Shizhao Sun, Zijiang James Yang, Jian-Guang Lou, and Dongmei Zhang. Layout- Prompter: Awaken the design ability of large language models.arXiv preprint arXiv:2311.06495, 2023. URLhttps://arxiv.org/abs/2311.06495

    arXiv 2023

  23. [23]

    G-Eval: NLG evaluation using GPT-4 with better human alignment.arXiv preprint arXiv:2303.16634, 2023

    Yang Liu, Dan Iter, Yichong Xu, Shuohang Wang, Ruochen Xu, and Chenguang Zhu. G-Eval: NLG evaluation using GPT-4 with better human alignment.arXiv preprint arXiv:2303.16634, 2023. URL https://arxiv.org/abs/2303.16634

    Pith/arXiv arXiv 2023

  24. [24]

    Self-refine: Iterative refinement with self-feedback

    Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, et al. Self-refine: Iterative refinement with self-feedback. InAdvances in Neural Information Processing Systems, volume 36, pages 46534–46594, 2023. URL https://proceedings.neurips.cc/paper_files/paper/2023/hash/ 9...

    2023

  25. [25]

    Presentations are not always linear! GNN meets LLM for text document-to-presentation transformation with attribution

    Himanshu Maheshwari, Sambaran Bandyopadhyay, Aparna Garimella, and Anandhavelu Natarajan. Presentations are not always linear! GNN meets LLM for text document-to-presentation transformation with attribution. InFindings of the Association for Computational Linguistics: EMNLP 2024, pages 15948–15962, Miami, Florida, USA, November 2024. Association for Compu...

    work page doi:10.18653/v1/2024.findings-emnlp.936 2024

  26. [26]

    Presentations by the humans and for the humans: Harnessing LLMs for generating persona-aware slides from documents

    Ishani Mondal, Shwetha S, Anandhavelu Natarajan, Aparna Garimella, Sambaran Bandyopadhyay, and Jordan Boyd-Graber. Presentations by the humans and for the humans: Harnessing LLMs for generating persona-aware slides from documents. InProceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long ...

    work page doi:10.18653/v1/2024.eacl-long.163 2024

  27. [28]

    URLhttps://arxiv.org/abs/2311.09180

    arXiv

  28. [29]

    WebGPT: Browser-assisted question-answering with human feedback.arXiv preprint arXiv:2112.09332, 2021

    Reiichiro Nakano, Jacob Hilton, Suchir Balaji, Jeff Wu, Long Ouyang, Christina Kim, Christopher Hesse, Shantanu Jain, Vineet Kosaraju, William Saunders, et al. WebGPT: Browser-assisted question-answering with human feedback.arXiv preprint arXiv:2112.09332, 2021. URL https://arxiv.org/abs/2112. 09332

    Pith/arXiv arXiv 2021

  29. [30]

    Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy V . V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mahmoud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Hervé Jégou, Julien Mairal, Patric...

    2024

  30. [31]

    Tarik Can Ozden, Sachidanand VS, Furkan Horoz, Ozgur Kara, Junho Kim, and James M. Rehg. Narrative- driven paper-to-slide generation via ArcDeck.arXiv preprint arXiv:2604.11969, 2026. doi: 10.48550/ arXiv.2604.11969. URLhttps://arxiv.org/abs/2604.11969

    Pith/arXiv arXiv 2026

  31. [32]

    MemGPT: Towards LLMs as Operating Systems

    Charles Packer, Sarah Wooders, Kevin Lin, Vivian Fang, Shishir G. Patil, Ion Stoica, and Joseph E. Gonzalez. Memgpt: Towards LLMs as operating systems.arXiv preprint arXiv:2310.08560, 2023. doi: 10.48550/arXiv.2310.08560. URLhttps://arxiv.org/abs/2310.08560

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2310.08560 2023

  32. [33]

    O’Brien, Carrie J

    Joon Sung Park, Joseph C. O’Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. Generative agents: Interactive simulacra of human behavior.arXiv preprint arXiv:2304.03442,

    Pith/arXiv arXiv

  33. [34]

    Generative Agents: Interactive Simulacra of Human Behavior

    doi: 10.48550/arXiv.2304.03442. URLhttps://arxiv.org/abs/2304.03442

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2304.03442

  34. [35]

    Patil, Tianjun Zhang, Xin Wang, and Joseph E

    Shishir G. Patil, Tianjun Zhang, Xin Wang, and Joseph E. Gonzalez. Gorilla: Large language model connected with massive APIs.arXiv preprint arXiv:2305.15334, 2023. URL https://arxiv.org/abs/ 2305.15334

    Pith/arXiv arXiv 2023

  35. [36]

    Bigham, and Jason Wu

    Yi-Hao Peng, Jeffrey P. Bigham, and Jason Wu. Designpref: Capturing personal preferences in visual design generation.arXiv preprint arXiv:2511.20513, 2025. doi: 10.48550/arXiv.2511.20513. URL https://arxiv.org/abs/2511.20513

    work page doi:10.48550/arxiv.2511.20513 2025

  36. [37]

    Rossi, Fan Du, Sungchul Kim, Eunyee Koh, Sana Malik, Tak Yeon Lee, and Nesreen K

    Xin Qian, Ryan A. Rossi, Fan Du, Sungchul Kim, Eunyee Koh, Sana Malik, Tak Yeon Lee, and Nesreen K. Ahmed. Personalized visualization recommendation.ACM Transactions on the Web, 16(3):11:1–11:47,

  37. [38]

    URLhttps://doi.org/10.1145/3538703

    doi: 10.1145/3538703. URLhttps://doi.org/10.1145/3538703

    work page doi:10.1145/3538703

  38. [39]

    ToolLLM: Facilitating large language models to master 16000+ real-world APIs.arXiv preprint arXiv:2307.16789, 2023

    Yujia Qin, Shihao Liang, Yining Ye, Kunlun Zhu, Lan Yan, Yaxi Lu, Yankai Lin, Xin Cong, Xiangru Tang, Bill Qian, Sihan Zhao, Lauren Hong, Runchu Tian, Ruobing Xie, Jie Zhou, Mark Gerstein, Dahai Li, Zhiyuan Liu, and Maosong Sun. ToolLLM: Facilitating large language models to master 16000+ real-world APIs.arXiv preprint arXiv:2307.16789, 2023. URLhttps://a...

    Pith/arXiv arXiv 2023

  39. [40]

    LaMP: When large language models meet personalization.arXiv preprint arXiv:2304.11406, 2023

    Alireza Salemi, Sheshera Mysore, Michael Bendersky, and Hamed Zamani. LaMP: When large language models meet personalization.arXiv preprint arXiv:2304.11406, 2023. doi: 10.48550/arXiv.2304.11406. URLhttps://arxiv.org/abs/2304.11406

    work page doi:10.48550/arxiv.2304.11406 2023

  40. [41]

    Toolformer: Language models can teach themselves to use tools

    Timo Schick, Jane Dwivedi-Yu, Roberto Dessì, Roberta Raileanu, Maria Lomeli, Eric Hambro, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. Toolformer: Language models can teach themselves to use tools. InAdvances in Neural Information Processing Systems, volume 36, pages 68539–68551, 2023. URL https://proceedings.neurips.cc/paper_files/paper/2023/ha...

    2023

  41. [42]

    Reflexion: Language agents with verbal reinforcement learning

    Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. Reflexion: Language agents with verbal reinforcement learning. InAdvances in Neural Information Processing Systems, volume 36, pages 8634–8652, 2023. URL https://proceedings.neurips.cc/paper_ files/paper/2023/hash/1b44b878bb782e6954cd888628510e90-Abstract-Conference.html. 12

    2023

  42. [43]

    Edward Sun, Yufang Hou, Dakuo Wang, Yunfeng Zhang, and Nancy X. R. Wang. D2S: Document- to-slide generation via query-based text summarization.arXiv preprint arXiv:2105.03664, 2021. doi: 10.48550/arXiv.2105.03664. URLhttps://arxiv.org/abs/2105.03664

    work page doi:10.48550/arxiv.2105.03664 2021

  43. [44]

    LayoutNUWA: Revealing the hidden layout expertise of large language models.arXiv preprint arXiv:2309.09506, 2023

    Zecheng Tang, Chenfei Wu, Juntao Li, and Nan Duan. LayoutNUWA: Revealing the hidden layout expertise of large language models.arXiv preprint arXiv:2309.09506, 2023. URL https://arxiv.org/ abs/2309.09506

    arXiv 2023

  44. [45]

    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. URLhttps://arxiv.org/abs/2305.16291

    Pith/arXiv arXiv 2023

  45. [47]

    Le, Ed H

    Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc V . Le, Ed H. Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. Self-consistency improves chain of thought reasoning in language models.arXiv preprint arXiv:2203.11171, 2022. URLhttps://arxiv.org/abs/2203.11171

    Pith/arXiv arXiv 2022

  46. [48]

    MIRIX: Multi-Agent Memory System for LLM-Based Agents

    Yu Wang and Xi Chen. Mirix: Multi-agent memory system for LLM-based agents.arXiv preprint arXiv:2507.07957, 2025. doi: 10.48550/arXiv.2507.07957. URL https://arxiv.org/abs/2507. 07957

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2507.07957 2025

  47. [49]

    AutoGen: Enabling next-gen LLM applications via multi-agent conversa- tion.arXiv preprint arXiv:2308.08155, 2023

    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 conversa- tion.arXiv preprint arXiv:2308.08155, 2023. URLhttps://arxiv.org/abs/2308.08155

    Pith/arXiv arXiv 2023

  48. [50]

    From human memory to AI memory: A survey on memory mechanisms in the era of LLMs

    Yaxiong Wu, Sheng Liang, Chen Zhang, Yichao Wang, Yongyue Zhang, Huifeng Guo, Ruiming Tang, and Yong Liu. From human memory to AI memory: A survey on memory mechanisms in the era of LLMs. arXiv preprint arXiv:2504.15965, 2025. URLhttps://arxiv.org/abs/2504.15965

    Pith/arXiv arXiv 2025

  49. [51]

    Personalized response generation via generative split memory network

    Yuwei Wu, Xuezhe Ma, and Diyi Yang. Personalized response generation via generative split memory network. InProceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1956–1970, Online, June 2021. Association for Computational Linguistics. doi: 10.18653/v1/2021.naac...

    work page doi:10.18653/v1/2021.naacl-main.157 2021

  50. [52]

    SlideBot: A multi-agent framework for generating informative, reliable, multi-modal presentations.arXiv preprint arXiv:2511.09804, 2025

    Eric Xie, Danielle Waterfield, Michael Kennedy, and Aidong Zhang. SlideBot: A multi-agent framework for generating informative, reliable, multi-modal presentations.arXiv preprint arXiv:2511.09804, 2025. doi: 10.48550/arXiv.2511.09804. URLhttps://arxiv.org/abs/2511.09804

    work page doi:10.48550/arxiv.2511.09804 2025

  51. [53]

    A-MEM: Agentic Memory for LLM Agents

    Wujiang Xu, Zujie Liang, Kai Mei, Hang Gao, Juntao Tan, and Yongfeng Zhang. A-MEM: Agentic memory for LLM agents.arXiv preprint arXiv:2502.12110, 2025. doi: 10.48550/arXiv.2502.12110. URL https://arxiv.org/abs/2502.12110

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2502.12110 2025

  52. [54]

    PreGenie: An agentic framework for high-quality visual presentation generation

    Xiaojie Xu, Xinli Xu, Sirui Chen, Haoyu Chen, Fan Zhang, and Ying-Cong Chen. PreGenie: An agentic framework for high-quality visual presentation generation. InFindings of the Association for Computational Linguistics: EMNLP 2025, pages 3045–3063, Suzhou, China, November 2025. As- sociation for Computational Linguistics. doi: 10.18653/v1/2025.findings-emnl...

    work page doi:10.18653/v1/2025.findings-emnlp.165 2025

  53. [56]

    URLhttps://arxiv.org/abs/2305.10601

    Pith/arXiv arXiv

  54. [57]

    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. InThe Eleventh International Conference on Learning Representations, 2023. URLhttps://openreview.net/forum?id=WE_vluYUL-X

    2023

  55. [58]

    Agentic Memory: Learning Unified Long-Term and Short-Term Memory Management for Large Language Model Agents

    Yi Yu, Liuyi Yao, Yuexiang Xie, Qingquan Tan, Jiaqi Feng, Yaliang Li, and Libing Wu. Agentic memory: Learning unified long-term and short-term memory management for large language model agents.arXiv preprint arXiv:2601.01885, 2026. doi: 10.48550/arXiv.2601.01885. URL https://arxiv.org/abs/ 2601.01885

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2601.01885 2026

  56. [59]

    Fast-decoding diffusion language models via progress-aware confidence schedules.arXiv preprint arXiv:2512.02892, 2025

    Wenzheng Zeng, Mingyu Ouyang, Langyuan Cui, and Hwee Tou Ng. SlideTailor: Personalized presentation slide generation for scientific papers.arXiv preprint arXiv:2512.20292, 2025. doi: 10.48550/arXiv.2512. 20292. URLhttps://arxiv.org/abs/2512.20292. 13

    work page doi:10.48550/arxiv.2512 2025

  57. [61]

    URLhttps://arxiv.org/abs/2303.11589

    arXiv

  58. [62]

    Zhehao Zhang, Ryan A. Rossi, Branislav Kveton, Yijia Shao, Diyi Yang, Hamed Zamani, Franck Der- noncourt, Joe Barrow, Tong Yu, Sungchul Kim, Ruiyi Zhang, Jiuxiang Gu, Tyler Derr, Hongjie Chen, Junda Wu, Xiang Chen, Zichao Wang, Subrata Mitra, Nedim Lipka, Nesreen Ahmed, and Yu Wang. Personalization of large language models: A survey.arXiv preprint arXiv:2...

    arXiv 2024

  59. [63]

    PPTAgent: Generating and evaluating presentations beyond text-to- slides

    Hao Zheng, Xinyan Guan, Hao Kong, Wenkai Zhang, Jia Zheng, Weixiang Zhou, Hongyu Lin, Yaojie Lu, Xianpei Han, and Le Sun. PPTAgent: Generating and evaluating presentations beyond text-to- slides. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 14402–14418, Suzhou, China, November 2025. Association for Comput...

    work page doi:10.18653/v1/2025.emnlp-main.728 2025

  60. [64]

    DeepPresenter: Environment-Grounded Reflection for Agentic Presentation Generation

    Hao Zheng, Guozhao Mo, Xinru Yan, Qianhao Yuan, Wenkai Zhang, Xuanang Chen, Yaojie Lu, Hongyu Lin, Xianpei Han, and Le Sun. DeepPresenter: Environment-grounded reflection for agentic presentation generation.arXiv preprint arXiv:2602.22839, 2026. doi: 10.48550/arXiv.2602.22839. URL https: //arxiv.org/abs/2602.22839

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2602.22839 2026

  61. [65]

    Xing, et al

    Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric P. Xing, et al. Judging LLM-as-a-judge with MT-bench and chatbot arena. arXiv preprint arXiv:2306.05685, 2023. URLhttps://arxiv.org/abs/2306.05685

    Pith/arXiv arXiv 2023

  62. [66]

    MemoryBank: Enhancing Large Language Models with Long-Term Memory

    Wanjun Zhong, Lianghong Guo, Qiqi Gao, He Ye, and Yanlin Wang. Memorybank: Enhancing large language models with long-term memory.arXiv preprint arXiv:2305.10250, 2023. doi: 10.48550/arXiv. 2305.10250. URLhttps://arxiv.org/abs/2305.10250

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv 2023

  63. [67]

    objective

    Lianghui Zhu, Xinggang Wang, and Xinlong Wang. JudgeLM: Fine-tuned large language models are scalable judges.arXiv preprint arXiv:2310.17631, 2023. URL https://arxiv.org/abs/2310.17631. 14 A Appendix The appendix provides protocol details, supplemental quantitative tables, and qualitative examples that clarify how different memory signals appear in genera...

    arXiv 2023

  64. [68]

    Management Analyst Blue minimalist business template Template Slide Generated Slide

  65. [69]

    Marketing Manager Energetic blue-yellow template Template Slide Generated Slide

  66. [70]

    If a new slide is added later,its title should also be blue

    Public Policy Maker Formal red civic template Template Slide Generated Slide Source paper: Attention Is All You Need. Template slides use the manually edited images in template_previews/. Figure 7: Template-guided generation examples: selected template slides (top) and matched per- sona/template generations (bottom). A.8 Tool-Memory Pair-Level Details The...