arxiv: 2604.08491 · v2 · submitted 2026-04-09 · 💻 cs.HC

Recognition: unknown

Figures as Interfaces: Toward LLM-Native Artifacts for Scientific Discovery

Yifang Wang , Rui Sheng , Erzhuo Shao , Yifan Qian , Haotian Li , Nan Cao , Dashun Wang

Authors on Pith no claims yet

Pith reviewed 2026-05-10 17:11 UTC · model grok-4.3

classification 💻 cs.HC

keywords LLM-native figuresdata provenancescientific visualizationhuman-AI interactionreproducibilitybidirectional interfacesscience of science

0 comments

The pith

LLM-native figures embed full provenance to let models and users trace and extend scientific analyses.

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

This paper introduces LLM-native figures as artifacts that combine human readability with machine addressability by embedding data, code, operations, and specs. A bidirectional interface allows LLMs to interpret figures not as images but as entry points to their generative process. Demonstrations in science of science show faster discovery cycles, higher reproducibility, and clearer reasoning paths between agents. If correct, this changes figures from final outputs into living interfaces that support iterative collaboration without losing context.

Core claim

LLM-native figures are data-driven artifacts that are simultaneously human-legible and machine-addressable, each embedding complete provenance including the data subset, analytical operations and code, and visualization specification used to generate it. This enables an LLM to trace selections back to their sources, generate code to extend analyses, and orchestrate new visualizations through natural-language instructions or direct manipulation via a hybrid language-visual interface with bidirectional mapping. In the science of science domain, this leads to accelerated discovery, improved reproducibility, and transparent reasoning across agents and users, redefining the figure as an interface

What carries the argument

The hybrid language-visual interface integrating LLM agents with a bidirectional mapping between figures and their underlying data, code, and specifications.

If this is right

Figures can serve as starting points for new analyses instead of endpoints.
Provenance embedding makes all analytical steps explicitly accessible to LLMs.
Natural language instructions suffice to modify or extend visualizations.
Multi-agent systems benefit from shared, context-rich artifacts.
Reproducibility gains follow from direct links to executable code and data.

Where Pith is reading between the lines

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

Similar native artifacts could be developed for other research outputs such as tables or simulation results.
Adoption might require new standards for figure formats that support provenance embedding.
This framework could integrate with existing visualization libraries to make the transition easier.
Long-term, it may influence how scientific papers are structured around interactive rather than static elements.

Load-bearing premise

A reliable bidirectional mapping between the rendered figure and its underlying data and code can be maintained without errors or constant human correction during LLM interactions.

What would settle it

A test case in which an LLM is asked to modify a figure's analysis and the resulting output is checked for fidelity to the original data and code without external fixes.

Figures

Figures reproduced from arXiv: 2604.08491 by Dashun Wang, Erzhuo Shao, Haotian Li, Nan Cao, Rui Sheng, Yifang Wang, Yifan Qian.

**Figure 2.** Figure 2: The conceptual framework for iterative exploration in data-driven scientific dis [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 3.** Figure 3: Case study: university innovation landscape. (a) [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗

**Figure 4.** Figure 4: Computational evaluation results. 4 Discussion Figures have long served as the primary medium through which scientists interpret and communicate data. Yet, in most computational workflows, they remain detached from the analytic processes that generated them. The framework introduced in this paper demonstrates that this separation is not intrinsic. By embedding full provenance within visual artifacts, figur… view at source ↗

**Figure 5.** Figure 5: System design in Nexus. (a) User input types include natural language input and graphical input. (b) The Hybrid User Interface Module serves as the frontend of the system. Left: natural language interaction using a traditional chatbot and an inline chatbot. Right: graphical interaction using interactive dashboards (e.g., click, brush, and hover). (c) The Data Management Module manages SciSci domain dataset… view at source ↗

read the original abstract

Large language models (LLMs) are transforming scientific workflows, not only through their generative capabilities but also through their emerging ability to use tools, reason about data, and coordinate complex analytical tasks. Yet in most human-AI collaborations, the primary outputs, figures, are still treated as static visual summaries: once rendered, they are handled by both humans and multimodal LLMs as images to be re-interpreted from pixels or captions. The emergent capabilities of LLMs open an opportunity to fundamentally rethink this paradigm. In this paper, we introduce the concept of LLM-native figures: data-driven artifacts that are simultaneously human-legible and machine-addressable. Unlike traditional plots, each artifact embeds complete provenance: the data subset, analytical operations and code, and visualization specification used to generate it. As a result, an LLM can "see through" the figure--tracing selections back to their sources, generating code to extend analyses, and orchestrating new visualizations through natural-language instructions or direct manipulation. We implement this concept through a hybrid language-visual interface that integrates LLM agents with a bidirectional mapping between figures and underlying data. Using the science of science domain as a testbed, we demonstrate that LLM-native figures can accelerate discovery, improve reproducibility, and make reasoning transparent across agents and users. More broadly, this work establishes a general framework for embedding provenance, interactivity, and explainability into the artifacts of modern research, redefining the figure not as an end product, but as an interface for discovery. For more details, please refer to the demo video available at www.llm-native-figure.com.

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.

Desk Editor's Note private letter to a colleague

The paper frames figures as LLM-addressable artifacts with embedded provenance, which is a fresh synthesis, but offers almost no evidence that the bidirectional mapping holds up under real LLM interactions.

read the letter

The central move is to stop treating rendered plots as static images and instead attach the full data subset, analysis code, and visualization spec so an LLM can trace selections, regenerate code, or spawn new views from natural-language instructions. The hybrid interface keeps the visual layer for humans while giving agents direct access underneath. That framing pulls together threads from interactive visualization and LLM tool use in a way that feels new for scientific artifacts, and the science-of-science testbed is a sensible place to try it because those plots often pull from messy, multi-source data that needs frequent reworking. The demo presumably shows the basic loop working in hand-crafted cases, which at least illustrates the intended flow without obvious contradictions in the setup. The soft spot is the missing validation. The abstract claims faster discovery, better reproducibility, and transparent reasoning, yet supplies no metrics, no baseline comparisons, and no tests of what breaks when an LLM issues an ambiguous command, drifts a data subset, or produces a visualization spec that no longer matches the rendered figure. Without those checks the bidirectional mapping stays an assumption rather than a demonstrated property. Readers working on AI-assisted research tools or visualization systems would find the concept useful to think with, even if they have to supply their own robustness experiments. I would send it to peer review because the idea is coherent and points to a real workflow gap; referees could usefully press for the error analysis and quantitative results that are currently absent.

Referee Report

1 major / 1 minor

Summary. The paper introduces LLM-native figures as data-driven artifacts that embed complete provenance (data subsets, code, analytical operations, and visualization specs) to enable bidirectional interaction between humans, LLMs, and the underlying data. Unlike static plots, these figures support tracing, code generation, and orchestration via natural-language instructions through a hybrid language-visual interface. Using a science-of-science testbed, the work claims to demonstrate accelerated discovery, improved reproducibility, and transparent reasoning across agents and users, establishing a general framework for provenance-embedded research artifacts.

Significance. If the bidirectional mapping can be shown to be robust, the framework could meaningfully advance human-AI scientific collaboration by redefining figures as active interfaces rather than passive outputs, with potential benefits for reproducibility and multi-agent workflows in data-intensive fields.

major comments (1)

[Abstract / testbed demonstration] Abstract and science-of-science testbed demonstration: the central claim that LLM-native figures accelerate discovery, improve reproducibility, and make reasoning transparent rests on an asserted demonstration, yet the manuscript provides no quantitative metrics, error rates, or systematic analysis of failure modes (e.g., ambiguous instructions, data-subset drift after LLM edits, or visualization-spec inconsistencies). This leaves the reliability of the bidirectional mapping unverified and is load-bearing for the primary contribution.

minor comments (1)

[Abstract] The demo video link is referenced but the manuscript text does not include sufficient standalone description of the interface mechanics or example interaction traces to allow readers to assess the hybrid mapping without external resources.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address the single major comment below, acknowledging where the current manuscript falls short and outlining specific revisions.

read point-by-point responses

Referee: [Abstract / testbed demonstration] Abstract and science-of-science testbed demonstration: the central claim that LLM-native figures accelerate discovery, improve reproducibility, and make reasoning transparent rests on an asserted demonstration, yet the manuscript provides no quantitative metrics, error rates, or systematic analysis of failure modes (e.g., ambiguous instructions, data-subset drift after LLM edits, or visualization-spec inconsistencies). This leaves the reliability of the bidirectional mapping unverified and is load-bearing for the primary contribution.

Authors: We agree that the science-of-science testbed demonstration is primarily qualitative and illustrative, consisting of worked examples that show provenance tracing, code generation, and natural-language orchestration rather than controlled quantitative evaluation. The manuscript does not report metrics such as task success rates, error rates, or systematic failure-mode analysis for issues like ambiguous instructions or post-edit data drift. This is a substantive limitation for claims about acceleration, reproducibility, and transparency. In the revised version we will add a dedicated evaluation subsection that (1) defines a set of representative tasks with success/failure criteria, (2) reports quantitative results (e.g., success rates over repeated trials, latency, and reproducibility scores), and (3) provides a categorized analysis of observed failure modes together with mitigation strategies. We will also make the testbed artifacts and query logs available to support reproducibility of the evaluation itself. revision: yes

Circularity Check

0 steps flagged

No significant circularity: conceptual framework with independent implementation and demonstration.

full rationale

The paper proposes LLM-native figures as a new class of artifacts embedding provenance for bidirectional LLM-human interaction. Its central claims rest on a definitional introduction of the concept, description of a hybrid language-visual interface, and empirical demonstration within a science-of-science testbed rather than any derivation chain. No equations, fitted parameters, or load-bearing self-citations appear that would reduce the claims to inputs by construction; the bidirectional mapping is presented as an implemented capability rather than a result derived from prior self-referential assumptions. The work is therefore self-contained against external benchmarks of reproducibility and transparency.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The work rests on the domain assumption that LLMs can reliably interpret and act on embedded provenance without systematic errors, plus the ad-hoc invention of the LLM-native figure construct itself.

axioms (1)

domain assumption LLMs can effectively use bidirectional mappings between rendered figures and underlying data/code for analysis extension
Invoked when claiming acceleration of discovery and transparent reasoning across agents.

invented entities (1)

LLM-native figure no independent evidence
purpose: Data-driven artifact that is both human-legible and machine-addressable with embedded provenance
Core new construct introduced to replace static plots; no independent evidence provided beyond the conceptual proposal.

pith-pipeline@v0.9.0 · 5602 in / 1264 out tokens · 41546 ms · 2026-05-10T17:11:06.122238+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

95 extracted references · 20 canonical work pages · 4 internal anchors

[1]

Anscombe, F. J. Graphs in statistical analysis.The american statistician27, 17–21 (1973)

1973
[2]

Introducing deep research (2025)

OpenAI. Introducing deep research (2025). URLhttps://openai.com/index/int roducing-deep-research/. Accessed: 2026-02-09

2025
[3]

Gemini deep research (2025)

Gemini, G. Gemini deep research (2025). URLhttps://gemini.google/overvi ew/deep-research/?hl=en. Accessed: 2026-02-09

2025
[4]

Transforming r&d with agentic ai: Introducing microsoft discovery (2025)

Datar, A. Transforming r&d with agentic ai: Introducing microsoft discovery (2025). URL https://azure.microsoft.com/en-us/blog/transforming-rd-with-a gentic-ai-introducing-microsoft-discovery/. Accessed: 2026-02-09

2025
[5]

Shao, E.et al.Sciscigpt: Advancing human-ai collaboration in the science of science.arXiv preprint arXiv:2504.05559(2025). 23

work page arXiv 2025
[6]

Tu, T.et al.Towards conversational diagnostic artificial intelligence.Nature(2025)

2025
[7]

Harvey Professional Class AI (2025)

AI, H. Harvey Professional Class AI (2025). URLhttps://www.harvey.ai/. Accessed: 2026-02-09

2025
[8]

Lu, C.et al.The ai scientist: Towards fully automated open-ended scientific discovery.arXiv preprint arXiv:2408.06292(2024)

work page internal anchor Pith review arXiv 2024
[9]

Creating an executable paper is a journey through open science.Communications Physics(2020)

Lasser, J. Creating an executable paper is a journey through open science.Communications Physics(2020)

2020
[10]

& Goulier, L

Konkol, M., N¨ ust, D. & Goulier, L. Publishing computational research-a review of infras- tructures for reproducible and transparent scholarly communication.Research integrity and peer review(2020)

2020
[11]

& Bora, A

Ziemann, M., Poulain, P. & Bora, A. The five pillars of computational reproducibility: bioinformatics and beyond.Briefings in Bioinformatics(2023)

2023
[12]

A publishing platform that places code front and centre (2025)

Nature. A publishing platform that places code front and centre (2025). URLhttps: //www.nature.com/articles/d41586-024-02577-1. Accessed: 2026-02-09

2025
[13]

Pioneering ’live-code’ article allows scientists to play with each other’s results (2025)

Nature. Pioneering ’live-code’ article allows scientists to play with each other’s results (2025). URLhttps://www.nature.com/articles/d41586-019-00724-7. Accessed: 2026-02-09

2025
[14]

InProceedings of the 2017 symposium on cloud computing, 405–418 (2017)

Pasquier, T.et al.Practical whole-system provenance capture. InProceedings of the 2017 symposium on cloud computing, 405–418 (2017)

2017
[15]

C., Arnold, C., Gur, Y

Rupprecht, L., Davis, J. C., Arnold, C., Gur, Y. & Bhagwat, D. Improving reproducibility of data science pipelines through transparent provenance capture.Proceedings of the VLDB Endowment(2020)

2020
[16]

Direct manipulation: A step beyond programming languages.Computer (1983)

Shneiderman, B. Direct manipulation: A step beyond programming languages.Computer (1983)

1983
[17]

Ware, C.Information visualization: perception for design(Morgan Kaufmann, 2019)

2019
[18]

InInformation visual- ization: Human-centered issues and perspectives, 154–175 (Springer, 2008)

Keim, D.et al.Visual analytics: Definition, process, and challenges. InInformation visual- ization: Human-centered issues and perspectives, 154–175 (Springer, 2008)

2008
[19]

Munzner, T.Visualization analysis and design(CRC press, 2014). 24

2014
[20]

Zheng, T.et al.From automation to autonomy: A survey on large language models in scientific discovery.arXiv preprint arXiv:2505.13259(2025)

work page arXiv 2025
[21]

InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing(Association for Computational Linguistics, 2024)

Zhang, Y.et al.A comprehensive survey of scientific large language models and their applications in scientific discovery. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing(Association for Computational Linguistics, 2024)

2024
[22]

Wang, H.et al.Scientific discovery in the age of artificial intelligence.Nature620, 47–60 (2023)

2023
[23]

ACM Computing Surveys57, 1–38 (2025)

Zhang, Q.et al.Scientific large language models: A survey on biological & chemical domains. ACM Computing Surveys57, 1–38 (2025)

2025
[24]

& Buehler, M

Ghafarollahi, A. & Buehler, M. J. SciAgents: automating scientific discovery through bioin- spired multi-agent intelligent graph reasoning.Advanced Materials(2025)

2025
[25]

Zheng, Y.et al.Large language models for scientific discovery in molecular property predic- tion.Nature Machine Intelligence1–11 (2025)

2025
[26]

Maojun, S.et al.A survey on large language model-based agents for statistics and data science.The American Statistician1–21 (2025)

2025
[27]

P.et al.Data analysis in the era of generative ai.arXiv preprint arXiv:2409.18475 (2024)

Inala, J. P.et al.Data analysis in the era of generative ai.arXiv preprint arXiv:2409.18475 (2024)

work page arXiv 2024
[28]

arXiv preprint arXiv:2412.14222(2024)

Sun, M.et al.A survey on large language model-based agents for statistics and data science. arXiv preprint arXiv:2412.14222(2024)

work page arXiv 2024
[29]

Hong, S.et al.Data interpreter: An llm agent for data science.arXiv preprint arXiv:2402.18679(2024)

work page arXiv 2024
[30]

Guo, S.et al.Ds-agent: Automated data science by empowering large language models with case-based reasoning.arXiv preprint arXiv:2402.17453(2024)

work page arXiv 2024
[31]

Wang, D.et al.Human-ai collaboration in data science: Exploring data scientists’ perceptions of automated ai.Proceedings of the ACM on human-computer interaction3, 1–24 (2019)

2019
[32]

S., Zhu, K

Manning, B. S., Zhu, K. & Horton, J. J. Automated social science: Language models as scientist and subjects. Tech. Rep., National Bureau of Economic Research (2024)

2024
[33]

Gao, S.et al.Democratizing ai scientists using ToolUniverse.arXiv preprint arXiv:2509.23426(2025). 25

work page arXiv 2025
[34]

Liang, Y.et al.SkillNet: Create, evaluate, and connect ai skills.arXiv preprint arXiv:2603.04448(2026)

work page arXiv 2026
[35]

Huang, K.et al.Biomni: A general-purpose biomedical AI agent.biorxiv(2025)

2025
[36]

Sphinx: Enabling data science across academia

Sphinx. Sphinx: Enabling data science across academia. (2026). URLhttps://www.sp hinx.ai. Accessed: 2026-02-09

2026
[37]

Introducing observable canvases a collaborative, visual, spatial medium for data analysis (2026)

Observable. Introducing observable canvases a collaborative, visual, spatial medium for data analysis (2026). URLhttps://observablehq.com/blog/introducing-can vases-early-access. Accessed: 2026-02-09

2026
[38]

Zhu, D.et al.PaperBanana: Automating academic illustration for ai scientists.arXiv preprint arXiv:2601.23265(2026)

work page arXiv 2026
[39]

Plottie: Free to explore, collect and inspire your next figure

Plottie. Plottie: Free to explore, collect and inspire your next figure. discover high-quality scientific plots from open-access literature. (2026). URLhttps://plottie.art/. Accessed: 2026-02-09

2026
[40]

Lu, C.et al.Towards end-to-end automation of ai research.Nature651, 914–919 (2026)

2026
[41]

& Wang, D

Wang, Y., Qian, Y., Qi, X., Cao, N. & Wang, D. Innovationinsights: A visual analytics approach for understanding the dual frontiers of science and technology.IEEE Transactions on Visualization and Computer Graphics30, 518–528 (2023)

2023
[42]

Wang, Y.et al.Funding the Frontier: Visualizing the broad impact of science and science funding (2025).2509.16323

work page arXiv 2025
[43]

Introducing Observable Canvases (2025)

Bostock, M. Introducing Observable Canvases (2025). URLhttps://observablehq .com/blog/introducing-canvases-early-access. Accessed: 2026-02-09

2025
[44]

& Lee, B

Wang, C., Thompson, J. & Lee, B. Data Formulator: Ai-powered concept-driven visualization authoring.IEEE Transactions on Visualization and Computer Graphics30, 1128–1138 (2023)

2023
[45]

M., Marshall, D

Wang, C., Lee, B., Drucker, S. M., Marshall, D. & Gao, J. Data Formulator 2: Iterative creation of data visualizations, with ai transforming data along the way. InProceedings of the 2025 CHI Conference on Human Factors in Computing Systems, 1–17 (2025)

2025
[46]

Tableau Agent (2025)

Tableau. Tableau Agent (2025). URLhttps://www.tableau.com/products/tab leau-agent. Accessed: 2026-02-09. 26

2025
[47]

IEEE Transactions on Visualization and Computer Graphics(2024)

Tian, Y.et al.Chartgpt: Leveraging llms to generate charts from abstract natural language. IEEE Transactions on Visualization and Computer Graphics(2024)

2024
[48]

& Wang, Y

Wang, L., Zhang, S., Wang, Y., Lim, E.-P. & Wang, Y. LLM4Vis: Explainable visualization recommendation using ChatGPT. In Wang, M. & Zitouni, I. (eds.)Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing

2023
[49]

LIDA: A tool for automatic generation of grammar-agnostic visualizations and infographics using large language models

Dibia, V. LIDA: A tool for automatic generation of grammar-agnostic visualizations and infographics using large language models. InProceedings of the 61st Annual Meeting of the Association for Computational Linguistics(Association for Computational Linguistics, 2023)

2023
[50]

Zhao, Y.et al.LightV A: Lightweight visual analytics with llm agent-based task planning and execution.IEEE Transactions on Visualization and Computer Graphics(2024)

2024
[51]

Zhao, Y.et al.LA V A: Using large language models to enhance visual analytics.IEEE transactions on visualization and computer graphics(2024)

2024
[52]

Lange, D.et al.YAC: Bridging natural language and interactive visual exploration with generative ai for biomedical data discovery.arXiv preprint arXiv:2509.19182(2025)

work page arXiv 2025
[53]

Zhao, Y.et al.ProactiveV A: Proactive visual analytics with llm-based ui agent.arXiv preprint arXiv:2507.18165(2025)

work page arXiv 2025
[54]

In Extended Abstracts of the CHI Conference on Human Factors in Computing Systems(2024)

Gao, J.et al.A taxonomy for human-llm interaction modes: An initial exploration. In Extended Abstracts of the CHI Conference on Human Factors in Computing Systems(2024)

2024
[55]

Shen, L., Li, H., Wang, Y., Xie, X. & Qu, H. Prompting generative ai with interaction- augmented instructions. InExtended Abstracts of the CHI Conference on Human Factors in Computing Systems, 1–9 (2025)

2025
[56]

& Gadiraju, U

He, G., Demartini, G. & Gadiraju, U. Plan-then-execute: An empirical study of user trust and team performance when using llm agents as a daily assistant. InProceedings of the 2025 CHI Conference on Human Factors in Computing Systems(2025)

2025
[57]

Shen, L., Wang, Y., Qu, H., Xie, X. & Li, H. Interaction-augmented instruction: Modeling the synergy of prompts and interactions in human-genai collaboration. InCHI 2026(2026)

2026
[58]

Luera, R.et al.Survey of user interface design and interaction techniques in generative ai applications.arXiv preprint arXiv:2410.223701–42 (2024). 27

work page arXiv 2024
[59]

Generative Interfaces for Language Models

Chen, J., Zhang, Y., Zhang, Y., Shao, Y. & Yang, D. Generative interfaces for language models.arXiv preprint arXiv:2508.19227(2025)

work page internal anchor Pith review Pith/arXiv arXiv 2025
[60]

ChatGPT Canvas (2025)

ChatGPT. ChatGPT Canvas (2025). URLhttps://openai.com/index/introdu cing-canvas/. Accessed: 2026-02-09

2025
[61]

What are artifacts and how do i use them? (2025)

Claude. What are artifacts and how do i use them? (2025). URLhttps://support.cl aude.com/en/articles/9487310-what-are-artifacts-and-how-do-i -use-them. Accessed: 2026-02-09

work page arXiv 2025
[62]

& Xia, H

Cao, Y., Jiang, P. & Xia, H. Generative and malleable user interfaces with generative and evolving task-driven data model. InProceedings of the 2025 CHI Conference on Human Factors in Computing Systems(2025)

2025
[63]

& Xia, H

Suh, S., Min, B., Palani, S. & Xia, H. Sensecape: Enabling multilevel exploration and sensemaking with large language models. InProceedings of the 36th annual ACM symposium on user interface software and technology, 1–18 (2023)

2023
[64]

You, W.et al.DesignManager: An agent-powered copilot for designers to integrate ai design tools into creative workflows.ACM Transactions on Graphics (TOG)(2025)

2025
[65]

Get to know BigQuery data canvas: an ai-centric experience to reimagine data analytics (2025)

BigQuery, G. Get to know BigQuery data canvas: an ai-centric experience to reimagine data analytics (2025). URLhttps://cloud.google.com/blog/products/data-a nalytics/get-to-know-bigquery-data-canvas. Accessed: 2026-02-09

2025
[66]

& Heer, J

Satyanarayan, A., Moritz, D., Wongsuphasawat, K. & Heer, J. Vega-lite: A grammar of interactive graphics.IEEE transactions on visualization and computer graphics23, 341–350 (2016)

2016
[67]

& Jones, B

Ahmadpoor, M. & Jones, B. F. The dual frontier: Patented inventions and prior scientific advance.Science357, 583–587 (2017)

2017
[68]

Liang, W., Elrod, S., McFarland, D. A. & Zou, J. Systematic analysis of 50 years of stanford university technology transfer and commercialization.Patterns3(2022)

2022
[69]

& Jones, B

Yin, Y., Dong, Y., Wang, K., Wang, D. & Jones, B. F. Public use and public funding of science.Nature human behaviour6, 1344–1350 (2022)

2022
[70]

Tripodi, G.et al.Tenure and research trajectories.Proceedings of the National Academy of Sciences122, e2500322122 (2025). 28

2025
[71]

InInternational Conference on Learning Representations (ICLR)(2023)

Yao, S.et al.React: Synergizing reasoning and acting in language models. InInternational Conference on Learning Representations (ICLR)(2023)

2023
[72]

Advances in neural information processing systems36, 11809–11822 (2023)

Yao, S.et al.Tree of thoughts: Deliberate problem solving with large language models. Advances in neural information processing systems36, 11809–11822 (2023)

2023
[73]

Schick, T.et al.Toolformer: Language models can teach themselves to use tools.Advances in Neural Information Processing Systems36, 68539–68551 (2023)

2023
[74]

Madaan, A.et al.Self-refine: Iterative refinement with self-feedback.Advances in Neural Information Processing Systems36, 46534–46594 (2023)

2023
[75]

& Yao, S

Shinn, N., Cassano, F., Gopinath, A., Narasimhan, K. & Yao, S. Reflexion: Language agents with verbal reinforcement learning.Advances in Neural Information Processing Systems36, 8634–8652 (2023)

2023
[76]

Zhuang, Y.et al.Toolchain*: Efficient action space navigation in large language models with a* search.arXiv preprint arXiv:2310.13227(2023)

work page arXiv 2023
[77]

Chen, Q.et al.Vizlinter: A linter and fixer framework for data visualization.IEEE transactions on visualization and computer graphics28, 206–216 (2021)

2021
[78]

Liu, X.et al.A survey of nl2sql with large language models: Where are we, and where are we going?arXiv preprint arXiv:2408.05109(2024)

work page arXiv 2024
[79]

Hong, Z.et al.Next-generation database interfaces: A survey of llm-based text-to-sql.IEEE Transactions on Knowledge and Data Engineering(2025)

2025
[80]

IEEE Transactions on Visualization and Computer Graphics28, 5049–5070 (2021)

Wu, A.et al.AI4VIS: Survey on artificial intelligence approaches for data visualization. IEEE Transactions on Visualization and Computer Graphics28, 5049–5070 (2021)

2021

Showing first 80 references.