Pith. sign in

REVIEW 3 major objections 5 minor 46 references

Metadata Management for AI-Augmented Data Workflows

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper argues that AI-augmented data workflows need a system-level metadata layer, and proposes TableVault, whose central claim is that every data artifact and operation should be recorded, with the records themselves usable as data arti

desk verdict A sensible design vision for metadata in AI-augmented workflows, with genuinely novel bits in TableReference and execution-parameter lineage, but the 'guaranteed' language outruns the file-system-based implementation. read the letter →

arxiv 2508.06814 v1 pith:GBGYBE2U submitted 2025-08-09 cs.DB

Jinjin Zhao , Sanjay Krishnan This is my paper
classification cs.DB
keywords LLMagentsETLsystemsdataprovenancemetadatamanagementdataframesdocumentretrievallineage-awarereferencesreproducibility
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that AI-augmented data workflows — where humans and language models jointly clean, label, and transform data — need metadata managed at the systems level, and it proposes TableVault as a concrete design for doing so. Its central claim is that the existence of every data artifact and every operation in such workflows should be recorded, and that the records themselves should be usable as data artifacts. TableVault brings database-style guarantees — ACID-like execution, enforced ingestion through a single API, archived metadata on deletion — to pipelines whose individual steps are often opaque model calls. If this holds, mixed human-model pipelines become auditable and reproducible: one could reconstruct which prompt, which documents, and which model call produced a given dataset, and what any human edits changed.

What carries the argument

The central mechanism is the TableReference string, a custom Pythonic string type that references repository instances, columns, and row ranges, including a 'self.index' relative form for row-wise operations. These strings implicitly trace lineage between table instances, can be recursive so one reference embeds another, and encode common execution patterns (reduction, one-to-one, cumulation, convolution, selection) in a single declarative form. Working with declarative builder YAML files, they unify operation inputs and execution parameters, so parameters can themselves point at repository data — which is how execution-parameter lineage is captured. The repository layout, tables as folders

What would settle it

Test the 'guaranteed record of ingestion' claim by writing a file directly into a repository's artifact folder from an external script or model, bypassing the TableVault API, and then querying the metadata layer: if the repository cannot flag the unrecorded write, attribute it to an author, or expose it in lineage queries, the guarantee fails. A systematic version would run a corpus of external edits, with and without author-supplied descriptions, and measure how often the recorded metadata still supports a faithful downstream reproduction of the resulting dataset.

Watch

Extended reading notes

Core claim

The paper's central claim is that every data artifact and operation in collaborative human-model workflows should be recorded, and the records themselves should be usable as data artifacts. TableVault adopts an 'opportunistic' approach: in-system operations get full lineage tracing; external operations are flagged by recorded ingestion events. Four properties carry the design: guaranteed ingestion records (all data enters through the API), robust operation-status tracing across pauses and restarts, lineage of execution parameters back to their data origins, and a metadata layer queryable inside model executions. Because language models transform data using internal knowledge, full lineage is

Load-bearing premise

The load-bearing premise is that every piece of data enters the repository through TableVault's own API, and that anything changed outside the system — such as the manual pandas edit in the case study — is truthfully described by the author; bypass the API or omit the description, and the 'guaranteed record' is only as good as what someone remembers to write down.

Editorial extensions

If this is right

  • A generated dataset's lineage stays reconstructible: which documents, prompts, and model calls produced it, and which human or process authored each operation.
  • Operations can be paused, reverted, or restarted mid-run, with every such decision recorded, so data seen by later stages always matches the final recorded operation status.
  • Agents inside a workflow can query the metadata layer programmatically, and those queries are themselves logged for later introspection.
  • Execution parameters chosen by data-driven tuning — prompts, thread counts, model choices — get lineage back to the instances that determined them.
  • Even partially observable environments remain governable: full tracing is required only inside the system, while external modifications are constrained by enforced, author-attributed ingestion records.

Reading between the lines

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

  • Because TableVault stores its records as human-readable YAML, code, and log files, ordinary version-control and diff tools could audit the repository directly — an affordance the paper does not exploit but that follows from the storage choice.
  • The recursive 'agent spawns agent' workflow the paper sketches is its real stress test: each generation piles more parent-process lineage onto the metadata, and whether that lineage stays faithful at depth is untested; running a multi-level generation tree and verifying child-table provenance would settle it.
  • Routing every byte through one governing API is a strong architectural bet; a natural generalization is a provenance overlay that captures the same records from notebooks and ad-hoc scripts without requiring them to execute inside the vault, trading the 'guaranteed' record for wider adoption.
  • The 'records as data artifacts' principle implies a model could be pointed at its own past executions to plan the next step of a workflow — the paper envisions self-organizing workflows but does not demonstrate a model actually consuming the metadata for planning.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper presents TableVault, a Python-based metadata governance framework for AI-augmented data workflows. TableVault stores data artifacts and metadata in a native file-system repository, exposes a declarative builder interface and TableReference strings for lineage-aware references, and claims four properties: guaranteed record of data ingestion, robust operation status tracing, lineage capture of execution parameters, and metadata querying within model execution. The design is illustrated with a document-classification case study that combines OpenAI API calls, manual pandas edits, and externally imported artifacts, plus a preliminary latency evaluation of basic operations.

Significance. If the system delivers on its claims, it would address a practically important gap: metadata management for mixed human-AI pipelines where prompts, execution parameters, and human edits all affect data lineage. The TableReference string design and the idea of making metadata directly accessible as data artifacts are useful contributions, as is the emphasis on operation status and partially observable execution. The paper is honest in its scope, noting that performance and scalability are future work and that dynamic recursive generation is only partially supported. However, the central 'guaranteed record' claim is not backed by any enforcement mechanism, and the evaluation is too thin to substantiate the broader robustness and reproducibility claims.

major comments (3)
  1. [§2.3, §2.1] The 'guaranteed record of data ingestion' is not actually guaranteed by the architecture. Section 2.1 states that all data is stored in human-readable files on the native file system, with no operating-system access control, filesystem monitoring, or checksum/diff verification described. Section 2.3 asserts that 'all data ingestion into the repository must occur through its API,' but this is an assumption rather than an enforced property. Any user or model agent with write access to the repository directory can modify dataframes or artifacts directly, bypassing the API and leaving no metadata. The case study in Section 3.4 relies on the user voluntarily calling create_instance(external=True) and providing a free-text description, which is not automatically verified. Thus the central claim in Section 2.4 that 'the existence of all data artifacts and operations ... should be recorded' hold
  2. [§4] The experimental evaluation is too limited to support the paper's broader claims. Figure 3 and Figure 4 report only three trials per configuration, with no error bars, confidence intervals, or statistical analysis. There is no comparison against existing metadata or provenance systems (e.g., Ground, MLInspect, Smoke, which are cited in Section 5). The single document-classification case study is illustrative but does not exercise the claimed 'robust operation status tracing' under concurrency, interruption, or multi-agent scenarios. Section 3.5 describes a more complex recursive generation pattern but openly states it is 'partially supported' and requires user-provided metadata. Consequently, the Abstract's statement that TableVault enables 'robust metadata management, even in partially observable execution environments' is not substantiated by the presented evidence.
  3. [§1 vs. §2.3] The introduction presents a 'practical opportunistic approach' where full metadata is traced only for operations inside the system and external operations are signaled through recorded ingestion points. This is a reasonable design stance. However, Section 2.3 makes the stronger claim that TableVault 'prevents' unrecorded ingestion by controlling the storage layer. These two positions are inconsistent: the former accepts best-effort capture, while the latter asserts a guarantee. The paper should either implement an actual enforcement mechanism (e.g., filesystem permissions, a staging directory, or write-ahead logging for all file changes) or explicitly restate the guarantee as a usage policy that requires all writes to go through the API.
minor comments (5)
  1. [§5] Typo: 'Palimspzet' should be 'Palimpsest' (appears twice in the Related Work section). Also 'user0defined' should be 'user-defined'.
  2. [§4] Figures 3 and 4 are not self-contained: they lack axis labels explaining what is being measured and the units. The text says 'execution time of basic TableVault operations' but the specific operations and scales are not defined.
  3. [§2.3] The TableReference examples in Table 1 would benefit from a more formal grammar or BNF definition. The text explains the intuition but not the exact precedence and escaping rules, which are needed for the claims about 'universal' lineage capture.
  4. [§3.1] The code snippets use fictional module names and functions (e.g., 'create_paper_table_from_folder') but do not specify whether these are part of the released TableVault package or user-provided. Clarifying this would help reproducibility.
  5. [§2.4] The statement 'the records themselves should be directly accessible as data artifacts' is central, but the paper does not describe how metadata is physically represented as queryable data artifacts (e.g., whether they are dataframes themselves). A concrete example of metadata querying within model execution would strengthen the paper.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity; a minor non-load-bearing self-citation and a guarantee/implementation gap that is a robustness issue, not a circular derivation.

full rationale

The paper makes no quantitative predictions and has no fitted parameters; its central contribution is a normative design principle and a system architecture. The claim that TableVault records ingestion events, operation status, and parameter lineage is implemented by construction through API-managed operations and TableReference strings, which is a design choice rather than a result derived from inputs. The only self-citation is [46] (Zhao & Krishnan, ICDE 2024), cited for common data operation patterns (reduction, one-to-one, cumulation, convolution, selection) supporting the TableReference syntax; this citation is illustrative and not load-bearing for the metadata-governance claim, since the patterns are also enumerated in Table 1 of this paper. The more substantive issue—the 'guaranteed record of data ingestion' claim versus the native-filesystem repository and the external=True manual-edit path (Sections 2.3 and 3.4)—is a mismatch between stated guarantee and actual enforcement, not a circular argument: the paper does not fit a parameter to data and then rename it a prediction, and it does not import a uniqueness theorem from its own prior work. Thus circularity burden is low; score 2 reflects the minor self-citation, not any load-bearing circular step.

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

No new physical or mathematical entities are postulated. The TableReference string is a design artifact of the proposed system, not an entity invoked to explain observed phenomena; its existence is the contribution, not an assumption.

assumptions (4)
  • domain assumption A single overarching system can control top-level workflow orchestration and all data storage and versioning in a human-AI data workflow.
    This is the basis for the opportunistic capture strategy in Section 2; the system cannot deliver guaranteed metadata if users freely use external editors or storages.
  • domain assumption Data ingestion into the repository always occurs through the TableVault API.
    Stated in Section 2.3 ('All data ingestion into the repository must occur through its API'); the guaranteed ingestion record depends on this being enforceable, but nothing prevents a user from directly editing files.
  • domain assumption LLM execution is opaque, so full lineage cannot be traced by the execution system; linking to operation executions and relying on application-level explainability is sufficient.
    Section 2, 'Robust Operation Status Tracing'; this assumption defines the acceptable level of lineage, which is partial rather than complete.
  • ad hoc to paper The TableReference string and YAML builder files, when used, capture the data dependencies of operations accurately.
    Section 2.3 introduces these constructs; if a user provides a wrong reference, the captured lineage is wrong. The system does not validate semantic correctness of the declared dependencies.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Metadata Management for AI-Augmented Data Workflows." pith.science (2026). https://pith.science/paper/GBGYBE2U

@misc{pith2026250806814,
  author       = {Pith},
  title        = {Pith review of: Metadata Management for AI-Augmented Data Workflows},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GBGYBE2U}},
  note         = {Machine review of arXiv:2508.06814}
}
read the original abstract

AI-augmented data workflows introduce complex governance challenges, as both human and model-driven processes generate, transform, and consume data artifacts. These workflows blend heterogeneous tools, dynamic execution patterns, and opaque model decisions, making comprehensive metadata capture difficult. In this work, we present TableVault, a metadata governance framework designed for human-AI collaborative data creation. TableVault records ingestion events, traces operation status, links execution parameters to their data origins, and exposes a standardized metadata layer. By combining database-inspired guarantees with AI-oriented design, such as declarative operation builders and lineage-aware references, TableVault supports transparency and reproducibility across mixed human-model pipelines. Through a document classification case study, we demonstrate how TableVault preserves detailed lineage and operational context, enabling robust metadata management, even in partially observable execution environments.

Figures

Figures reproduced from arXiv: 2508.06814 by the authors.

Figure 1
Figure 1. System diagram of TableVault. of specialized tools with one system - for example, a generalized workflow system trying to beat a Jupyter notebook for quick ad￾hoc data transformations [21]. However, achieving comprehensive, end-to-end metadata capture across a heterogeneous toolchain is impracticable, as it would require deep instrumentation of every external component. Instead, we argue for a practical opportunisti… view at source ↗
Figure 2
Figure 2. Breakdown of a TableReference string. It has become common practice to parameterize language model operations with a builder YAML file [36]. While this doesn’t change system implementation, a key design decision in TableVault is that this file exposes execution parameters to the author, and unifies parametrization for execution and operations. This allows for com￾plex data-driven workflows, is that each of these par… view at source ↗
Figure 3
Figure 3. Execution time of basic TableVault operations. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Execution time of OpenAI response over documents. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

46 extracted references · 26 canonical work pages

  1. [1]

    Josh Abramson, Jonas Adler, Jack Dunger, Richard Evans, Tim Green, Alexander Pritzel, Olaf Ronneberger, Lindsay Willmore, Andrew J Ballard, Joshua Bambrick, Sebastian Bodenstein, David A Evans, Chia-Chun Hung, Michael O’Neill, David Metadata Management for AI-Augmented Data Workflows tbd. ’25, tbd., tbd/ Reiman, Kathryn Tunyasuvunakool, Zachary Wu, Akvil˙...

    work page 2024

  2. [3]

    Shah, Ben Sowell, Dan G

    Eric Anderson, Jonathan Fritz, Austin Lee, Bohou Li, Mark Lindblad, Henry Lindeman, Alex Meyer, Parth Parmar, Tanvi Ranade, Mehul A. Shah, Ben Sowell, Dan G. Tecuci, Vinayak Thapliyal, and Matt Welsh. 2024. The Design of an LLM-powered Unstructured Analytics System. ArXiv abs/2409.00847 (2024). https://api.semanticscholar.org/CorpusID:272368251

    arXiv 2024

  3. [4]

    Anthropic PBC. 2025. Model Context Protocol Specification . https:// modelcontextprotocol.io/specification/2025-06-18 Version 2025-06-18

    work page 2025

  4. [5]

    Simran Arora, Brandon Yang, Sabri Eyuboglu, Avanika Narayan, Andrew Ho- jel, Immanuel Trummer, and Christopher Ré. 2023. Language Models Enable Simple Systems for Generating Structured Views of Heterogeneous Data Lakes. Proc. VLDB Endow. 17 (2023), 92–105. https://api.semanticscholar.org/CorpusID: 258212828

    work page 2023

  5. [6]

    Shao-Qian Cai, Li qing Qiu, Xiaojun Chen, Qin Zhang, and Long Chen. 2022. Semantic-enhanced Image Clustering. ArXiv abs/2208.09849 (2022). https: //api.semanticscholar.org/CorpusID:251718842

    work page Pith review arXiv 2022

  6. [7]

    Harrison Chase and LangChain Inc. 2025. LangGraph: A framework for building stateful, long-running LLM agents . https://github.com/langchain-ai/langgraph Accessed 2025-08-04

    work page 2025

  7. [8]

    James Cheney, Laura Chiticariu, and Wang Chiew Tan. 2009. Provenance in Databases: Why, How, and Where. Found. Trends Databases 1 (2009), 379–474. https://api.semanticscholar.org/CorpusID:1778556

    work page 2009

  8. [9]

    Xin Luna Dong, Evgeniy Gabrilovich, Geremy Heitz, Wilko Horn, Kevin Murphy, Shaohua Sun, and Wei Zhang. 2015. From data fusion to knowledge fusion.arXiv preprint arXiv:1503.00302 (2015)

    work page Pith review arXiv 2015

Show all 46 references
  1. [10]

    Darren Edge, Ha Trinh, Newman Cheng, Joshua Bradley, Alex Chao, Apurva Mody, Steven Truitt, and Jonathan Larson. 2024. From Local to Global: A Graph RAG Approach to Query-Focused Summarization. ArXiv abs/2404.16130 (2024). https://api.semanticscholar.org/CorpusID:269363075

    2024 arXiv

  2. [11]

    Elmore, Michael J

    Raul Castro Fernandez, Aaron J. Elmore, Michael J. Franklin, Sanjay Krishnan, and Chenhao Tan. 2023. How Large Language Models Will Disrupt Data Man- agement. Proc. VLDB Endow. 16 (2023), 3302–3309. https://api.semanticscholar. org/CorpusID:261193780

    2023

  3. [12]

    Franklin, Donald Kossmann, Tim Kraska, Sukriti Ramesh, and Reynold Xin

    Michael J. Franklin, Donald Kossmann, Tim Kraska, Sukriti Ramesh, and Reynold Xin. 2011. CrowdDB: answering queries with crowdsourcing. In ACM SIGMOD Conference. https://api.semanticscholar.org/CorpusID:12809868

    2011

  4. [13]

    Stefan Grafberger, Julia Stoyanovich, and Sebastian Schelter. 2021. Lightweight Inspection of Data Preprocessing in Native Machine Learning Pipelines. In Con- ference on Innovative Data Systems Research . https://api.semanticscholar.org/ CorpusID:232203702

    2021

  5. [14]

    Jim Gray and Andreas Reuter. 1993. Transaction Processing: Concepts and Tech- niques. Morgan–Kaufmann, San Mateo, CA

    1993

  6. [15]

    Harris, K

    Charles R. Harris, K. Jarrod Millman, Stéfan J. van der Walt, Ralf Gommers, Pauli Virtanen, David Cournapeau, Eric Wieser, Julian Taylor, Sebastian Berg, Nathaniel J. Smith, Robert Kern, Matti Picus, Stephan Hoyer, Marten H. van Kerkwijk, Matthew Brett, Allan Haldane, Jaime Fe...

    2020

  7. [16]

    Joseph M Hellerstein, Vikram Sreekanti, Joseph E. Gonzalez, James Dalton, Akon Dey, Sreyashi Nag, Krishna Ramachandran, Sudhanshu Arora, Arka Bhat- tacharyya, Shirshanka Das, Mark Donsky, Gabriel Fierro, Chang She, Carl Steinbach, Venkat Ram Subramanian, and Eric Sun. 2017. Gr...

    2017

  8. [17]

    Mohammad Hosseini and Serge P.J.M. Horbach. 2023. Fighting reviewer fatigue or amplifying bias? Considerations and recommendations for use of ChatGPT and other large language models in scholarly peer review. Research Integrity and Peer Review 8 (2023). https://api.semanticscho...

    2023

  9. [18]

    Yaojie Hu, Ilias Fountalis, Jin Tian, and Nikolaos Vasiloglou. 2024. An- notatedTables: A Large Tabular Dataset with Language Model Annotations. arXiv:2406.16349 [cs.LG] https://arxiv.org/abs/2406.16349

    2024 arXiv

  10. [19]

    Gaurav Tarlok Kakkar, Jiashen Cao, Aubhro Sengupta, Joy Arulraj, and Hyesoon Kim. 2025. Aero: Adaptive Query Processing of ML Queries. Proceedings of the ACM on Management of Data 3 (2025), 1 – 27. https://api.semanticscholar.org/ CorpusID:279575403

    2025

  11. [20]

    Joshi, Hanna Moazam, Heather Miller, Matei Zaharia, and Christopher Potts

    Omar Khattab, Arnav Singhvi, Paridhi Maheshwari, Zhiyuan Zhang, Keshav Santhanam, Sri Vardhamanan, Saiful Haq, Ashutosh Sharma, Thomas T. Joshi, Hanna Moazam, Heather Miller, Matei Zaharia, and Christopher Potts. 2023. DSPy: Compiling Declarative Language Model Calls into Self...

    2023 arXiv

  12. [21]

    Thomas Kluyver, Benjamin Ragan-Kelley, Fernando Pérez, Brian Granger, Matthias Bussonnier, Jonathan Frederic, Kyle Kelley, Jessica Hamrick, Jason Grout, Sylvain Corlay, Paul Ivanov, Damián Avila, Safia Abdalla, and Carol Willing. 2016. Jupyter Notebooks – a publishing format f...

    2016 doi

  13. [22]

    Daniel Levine, Syed Asad Rizvi, Sacha Lévy, Nazreen Pallikkavaliyaveetil, David Zhang, Xingyu Chen, Sina Ghadermarzi, Ruiming Wu, Zihe Zheng, Ivan Vrkic, Anna Zhong, Daphne Raskin, Insu Han, Antonio Henrique de Oliveira Fonseca, Josue Ortega Caro, Amin Karbasi, Rahul Madhav Dh...

  14. [23]

    Chunwei Liu, Matthew Russo, Michael Cafarella, Lei Cao, Peter Baille Chen, Zui Chen, Michael Franklin, Tim Kraska, Samuel Madden, and Gerardo Vitagliano. 2024. A Declarative System for Optimizing AI Workloads. arXiv:2405.14696 [cs.CL] https://arxiv.org/abs/2405.14696

    2024 arXiv

  15. [24]

    Mingxuan Liu, Zhun Zhong, Jun Li, Gianni Franchi, Subhankar Roy, and Elisa Ricci. 2024. Organizing Unstructured Image Collections using Natural Lan- guage. ArXiv abs/2410.05217 (2024). https://api.semanticscholar.org/CorpusID: 273186686

    2024 arXiv

  16. [25]

    Zhi Liu. 2006. Reuter5050. UCI Machine Learning Repository. DOI: https://doi.org/10.24432/C5DS42

    2006 doi

  17. [26]

    Cafarella, Michael J

    Samuel Madden, Michael J. Cafarella, Michael J. Franklin, and Tim Kraska. 2024. Databases Unbound: Querying All of the World’s Bytes with AI. Proc. VLDB En- dow. 17 (2024), 4546–4554. https://api.semanticscholar.org/CorpusID:272727978

    2024

  18. [27]

    Ryan Marcus, Parimarjan Negi, Hongzi Mao, Nesime Tatbul, Mohammad Al- izadeh, and Tim Kraska. 2021. Bao: Making Learned Query Optimization Practical. In Proceedings of the 2021 ACM SIGMOD International Conference on Management of Data (SIGMOD ’21) . ACM, 1275–1288. doi:10.1145...

    2021

  19. [28]

    João Moura and the CrewAI contributors. 2025. CrewAI: A flexible multi-agent orchestration framework. CrewAI Inc. https://github.com/crewAIInc/crewAI Python package. Available on PyPI

    2025

  20. [29]

    The pandas development team. 2020. pandas-dev/pandas: Pandas. doi:10.5281/ zenodo.3509134

    2020

  21. [30]

    Liana Patel, Siddharth Jha, Melissa Pan, Harshit Gupta, Parth Asawa, Carlos Guestrin, and Matei Zaharia. 2025. Semantic Operators: A Declarative Model for Rich, AI-based Data Processing. arXiv:2407.11418 [cs.DB] https://arxiv.org/abs/ 2407.11418

    2025 arXiv

  22. [31]

    Gonzalez, Joseph M

    Devin Petersohn, Stephen Macke, Doris Xin, William Ma, Doris Lee, Xi- angxi Mo, Joseph E. Gonzalez, Joseph M. Hellerstein, Anthony D. Joseph, and Aditya Parameswaran. 2020. Towards Scalable Dataframe Systems. arXiv:2001.00888 [cs.DB] https://arxiv.org/abs/2001.00888

    2020 arXiv

  23. [32]

    Miller, and Sebastian Riedel

    Fabio Petroni, Tim Rocktäschel, Patrick Lewis, Anton Bakhtin, Yuxiang Wu, Alexander H. Miller, and Sebastian Riedel. 2019. Language Models as Knowledge Bases? ArXiv abs/1909.01066 (2019). https://api.semanticscholar.org/CorpusID: 202539551

    2019 arXiv

  24. [33]

    Fotis Psallidas and Eugene Wu. 2018. Smoke: Fine-grained Lineage at Interactive Speed. Proc. VLDB Endow. 11 (2018), 719–732. https://api.semanticscholar.org/ CorpusID:3591285

    2018

  25. [34]

    Matthew Russo, Sivaprasad Sudhir, Gerardo Vitagliano, Chunwei Liu, Tim Kraska, Samuel Madden, and Michael Cafarella. 2025. Abacus: A Cost-Based Optimizer for Semantic Operator Systems. arXiv:2505.14661 [cs.DB] https://arxiv.org/abs/ 2505.14661

    2025

  26. [35]

    Tobias Schomacker. 2018. Grimms’ Fairy Tales. Data set. Accessed 2025-08-04

    2018

  27. [36]

    Parameswaran, and Eugene Wu

    Shreya Shankar, Tristan Chambers, Tarak Shah, Aditya G. Parameswaran, and Eugene Wu. 2025. DocETL: Agentic Query Rewriting and Evaluation for Complex Document Processing. arXiv:2410.12189 [cs.DB] https://arxiv.org/abs/2410.12189

    2025 arXiv

  28. [37]

    Logan IV, Eric Wallace, and Sameer Singh

    Taylor Shin, Yasaman Razeghi, Robert L. Logan IV, Eric Wallace, and Sameer Singh. 2020. AutoPrompt: Eliciting Knowledge from Language Models with Automatically Generated Prompts. In Proc. of the 2020 Conf. on Empirical Methods in Natural Language Processing (EMNLP 2020) . 4222...

    2020 doi arXiv

  29. [38]

    Jasper Snoek, Hugo Larochelle, and Ryan P. Adams. 2012. Practical Bayesian Optimization of Machine Learning Algorithms. InAdvances in Neural Information Processing Systems 25 (NeurIPS 2012) . 2951–2959. https://arxiv.org/abs/1206.2944

    2012 arXiv

  30. [39]

    van Schaik and Brittany Pugh

    Tempest A. van Schaik and Brittany Pugh. 2024. A Field Guide to Automatic Evaluation of LLM-Generated Summaries. Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval tbd. ’25, tbd., tbd/ Jinjin Zhao and Sanjay Krishnan...

    2024

  31. [40]

    Vijay Viswanathan, Kiril Gashteovski, Carolin (Haas) Lawrence, Tong- shuang Sherry Wu, and Graham Neubig. 2023. Large Language Models Enable Few-Shot Clustering. Transactions of the Association for Computational Linguistics 12 (2023), 321–333. https://api.semanticscholar.org/C...

    2023

  32. [41]

    Junke Wang, Lingchen Meng, Zejia Weng, Bo He, Zuxuan Wu, and Yu-Gang Jiang. 2023. To See is to Believe: Prompting GPT-4V for Better Visual Instruction Tuning. arXiv:2311.07574 [cs.CV] https://arxiv.org/abs/2311.07574

    2023 arXiv

  33. [42]

    Wilkinson, Michel Dumontier, Iain J

    Mark D. Wilkinson, Michel Dumontier, Iain J. Aalbersberg, Gabrielle Appleton, Myles Axton, Arie Baak, Niklas Blomberg, Jan-Willem Boiten, Luiz Bonino da Silva Santos, Philip E. Bourne, Jildau Bouwman, Anthony J. Brookes, Timothy Clark, Mercè Crosas, Ingrid Dillo, Olivier Dumon...

    2016

  34. [43]

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. ReAct: Synergizing Reasoning and Acting in Language Models. arXiv:2210.03629 [cs.CL] https://arxiv.org/abs/2210.03629

    2023 arXiv

  35. [44]

    Qi Zeng, Mankeerat Sidhu, Hou Pong Chan, Lu Wang, and Heng Ji. 2023. Scientific Opinion Summarization: Paper Meta-review Generation Dataset, Methods, and Evaluation. In AI4Research/DemocrAI@IJCAI. https://api.semanticscholar.org/ CorpusID:258865977

    2023

  36. [45]

    Yuwei Zhang, Zihan Wang, and Jingbo Shang. 2023. ClusterLLM: Large Language Models as a Guide for Text Clustering. ArXiv abs/2305.14871 (2023). https: //api.semanticscholar.org/CorpusID:258866119

    2023 arXiv

  37. [46]

    Jinjin Zhao and Sanjay Krishnan. 2024. Compression and In-Situ Query Pro- cessing for Fine-Grained Array Lineage. In 40th IEEE International Conference on Data Engineering, ICDE 2024, Utrecht, The Netherlands, May 13-16, 2024 . IEEE, 3654–3667. doi:10.1109/ICDE60146.2024.00281

    2024

  38. [2024]

    bioRxiv (2024)

    Cell2Sentence: Teaching Large Language Models the Language of Biology. bioRxiv (2024). https://api.semanticscholar.org/CorpusID:262014469

    2024

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.