Pith. sign in

REVIEW 4 major objections 5 minor 39 references

GAIA is an agentic AI system that plans, retrieves, and executes geothermal field-development workflows.

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

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

T0 review · deepseek-v4-flash

2026-08-03 23:49 UTC pith:LQEX5HRF

load-bearing objection A clearly described agentic system for geothermal workflows, but the promised benchmark evaluation is completely absent—the paper's central claim is unsupported. the 4 major comments →

arxiv 2511.03852 v2 pith:LQEX5HRF submitted 2025-11-05 physics.geo-ph

Advancing Subsurface Discovery and Geothermal Monitoring with an Agentic Artificial Intelligence Framework

Randy Harsuko , Zhengfa Bi , Guodong Chen , Nori Nakata This is my paper
classification physics.geo-ph
keywords AI agentsgeothermal field developmentretrieval-augmented generationseismic monitoringphase pickingevent locationseismicity forecastingdigital twin
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper introduces GAIA, an agentic AI system meant to automate and assist geothermal field development. Its central claim is that a general-purpose language-model agent, paired with retrieval over a five-thousand-document geothermal knowledge base and a digital twin of classical seismic-analysis tools, can take a natural-language request and carry out the whole chain of monitoring tasks: phase picking, event location, magnitude estimation, and seismicity forecasting. The authors position GAIA as the first end-to-end agentic system for geothermal projects, arguing that such automation can compress multidisciplinary decision-making under time constraints. A curated benchmark is asserted, though the provided text demonstrates the workflows through four use cases rather than reporting benchmark scores.

Core claim

At the center is an agentic retrieval-augmented generation loop: a main agent, built on an open-weight instruction-tuned language model, thinks through the user's prompt, makes a step-by-step plan, and at each step either retrieves relevant context from a multi-modal vector database or executes a tool in GAIA's digital twin. The digital twin wraps standard seismological methods—STA/LTA and ML-based phase picking, Geiger location, Richter magnitude, ETAS forecasting—together with an automated visualization subagent, so the agent can select tools, set parameters, produce interactive plots, and formulate a final report. The paper claims this design automates repetitive seismic monitoring work,

What carries the argument

The load-bearing mechanism is the agentic retrieval-augmented generation loop: a main agent using chain-of-thought planning whose steps alternate between retrieving from a vector database of over five thousand geothermal papers plus images and tabular data, and executing tools exposed by the GAIA Digital Twin—a modular wrapper around standard seismological routines and automatic visualization-code generation. This loop is what converts a natural-language request into an executed multi-step workflow.

Load-bearing premise

The load-bearing premise is that a general-purpose instruction-tuned language model can reliably plan and select the right seismic-analysis tools without domain-specific fine-tuning, so that mistakes at the planning step do not cascade through the full pipeline.

What would settle it

Run GAIA on a published geothermal microseismic dataset with expert-reviewed phase picks, event locations, and magnitudes; ask it to produce an updated catalog, and compare outputs against the reference. If tool selection or parameter choices regularly yield locations or magnitudes outside physically plausible tolerances, or if pipeline errors compound at any stage, the central claim fails.

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

If this is right

  • Geothermal operators could request 'update the event catalog for the last 24 hours' and receive phase picks, locations, magnitudes, and updated forecasts without writing scripts.
  • Standard seismic tools become callable by a language model, so the agent's tool-selection and parameter choices are testable against expert practice.
  • The modular digital twin means adding new analysis methods (moment magnitude, double-difference location, deep-learning predictors) extends GAIA without redesigning the agent.
  • The same agentic pattern can be turned toward other stages of geothermal development the authors list—reservoir simulation, risk assessment, economic analysis—as tools are added.

Where Pith is reading between the lines

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

  • The paper's assertion of rigorous and continuous evaluation is supported in the visible text only by four illustrative use cases; benchmark questions and quantitative results are promised but not shown, so the evaluation claim is untested in this version.
  • A natural next experiment is to measure error rates at each stage of the pipeline separately (pick accuracy, location residuals, magnitude scatter, forecast skill), since errors at one stage propagate to the next.
  • Because the planner is a general language model with no geothermal fine-tuning, its planning decisions are the main risk; logging those decisions and comparing them with a human analyst's tool choices would isolate where failures originate.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper presents GAIA, an agentic retrieval-augmented generation (RAG) framework for geothermal field development. It combines a Gemma3-27B-IT main agent that plans and orchestrates multi-step workflows, a multimodal RAG knowledge base of over 5,000 documents embedded with Jina v4 and stored in LanceDB, and a 'digital twin' module that wraps classical seismological tools (STA/LTA, Geiger, Richter, ETAS) together with CodeGemma-based visualization. Four use cases are shown as qualitative, interactive-screenshot walkthroughs: phase picking, event location, magnitude estimation, and seismicity forecasting. The abstract and introduction claim that the system was rigorously and continuously evaluated on a curated benchmark set, but the full text does not contain such an evaluation.

Significance. If the benchmark claims were substantiated, GAIA would be a useful and arguably pioneering end-to-end agentic system for geothermal workflows. The modular architecture is clearly described, the choice of open-weight components (Gemma 3, Jina v4, ObsPy, Plotly, LanceDB) is practical and reproducible in principle, and the paper gives a helpful survey of related AI-agent work in geoscience. However, the actual contribution is currently an architectural demonstration, not a validated system: there are no metrics, baselines, error analyses, or comparisons against the classical tools that GAIA orchestrates. The central claim that the agentic orchestration is reliable enough for field development decision support is therefore unsupported by the evidence presented.

major comments (4)
  1. [§3 (opening paragraph) and Abstract] The abstract states that GAIA's performance was 'rigorously and continuously evaluated' on a curated benchmark set, and Section 3 announces: 'In the last subsection, we will also discuss the benchmarking of GAIA's performance using a curated test set.' No such subsection exists; the paper jumps from §3.4 to §4. There is no benchmark table, success rate, confidence interval, or comparison against the underlying STA/LTA, Geiger, Richter, or ETAS tools. This is not a presentation issue: the paper's central reliability claim is unsupported by the manuscript as written. The authors must either provide the promised benchmark data and analysis or explicitly withdraw/qualify the 'rigorously and continuously evaluated' claim.
  2. [§2.1 and Figure 1] The agent's Chain-of-Thought + ReAct planning and tool selection are assumed to work without domain fine-tuning. The paper provides no measurement of planning correctness, tool-selection accuracy, or retrieval quality. Because the workflow chains phase picking → location → magnitude → forecasting, an error at any step can propagate through the pipeline. No evidence is given about the frequency, magnitude, or consequences of such cascades. The benchmark should report step-level and end-to-end success/failure rates, with error analysis, not only qualitative examples.
  3. [§3.1–§3.4] The four use cases are illustrative walkthroughs with no input data, ground truth, or numerical output. For example, Figure 3 shows a phase-picking result but no comparison to known phase arrivals; Figures 4 and 5 show a location and magnitude estimate but no comparison to a catalog. Without quantitative validation, the reader cannot determine whether the displayed outputs are correct, typical, or cherry-picked. Quantitative results on the curated test set, including error bars and failure cases, are needed.
  4. [§2.1–§2.2 and overall manuscript] No code, data, benchmark, or configuration files are provided, and the composition of the 5,000-paper knowledge base is not described (sources, deduplication, document coverage, license). This blocks independent verification and makes the system irreproducible as presented. Even if a benchmark section is added, the results cannot be checked without a data/code release or at least a detailed benchmark specification.
minor comments (5)
  1. [Figure 4 caption] The caption says 'requests event magnitude estimation' but the use case in §3.2 is event location estimation; the text and figure should be made consistent.
  2. [§1, references] The citation 'Montes et al.' has no year or publication venue in the text, and the bibliography entry is incomplete. Please fix.
  3. [§2.3] 'Supports any file extensions' is immediately qualified by 'currently, we focus on support for structured tabular data'; the capability statement should be made precise to avoid overclaiming.
  4. [Title/headings] The full-text title appears as 'GAIA: ANAGENTICARTIFICIALINTELLIGENCESYSTEM FOR GEOTHERMALFIELDDEVELOPMENT' with spacing/formatting errors; please reconcile with the arXiv title.
  5. [§2.2] The term 'Digital Twin' is used in a broad sense for classical data-processing tools; consider defining this usage in the introduction or using a less loaded phrase such as 'tool module' to avoid confusion with conventional physics-based digital twins.

Circularity Check

0 steps flagged

No circularity: GAIA is a systems/architecture paper with no equations; self-citations are component references, not load-bearing reductions.

full rationale

Walking the claimed derivation chain, GAIA does not derive outputs from inputs in a way that reduces to definitions or fitted parameters. The abstract promises a curated benchmark and rigorous evaluation, and Section 3 announces 'In the last subsection, we will also discuss the benchmarking of GAIA's performance using a curated test set of geothermal-related questions,' but no such subsection appears; this is an omitted-evidence/internal-consistency problem rather than circularity. The four use cases (Sections 3.1-3.4) are demonstrations that classical tools (STA/LTA, Geiger, Richter, ETAS) are wrapped behind an LLM planner; there is no equation-level derivation. Self-citations to Nakata et al. [2025] for an ML phase-picking pretrained model, Nakata et al. [2024] for a planned real-time processor, and Bi et al. [2025] for planned deep-learning forecasting are references to components or future extensions, not to the paper's central orchestration claim. Even if one doubts the robustness of Gemma-3-based planning, that is an empirical/evidential concern, not a definitional or self-citational circularity. The benchmark claim is unverified, but unverified is not circular.

Axiom & Free-Parameter Ledger

2 free parameters · 3 axioms · 0 invented entities

No new physical entities are introduced; GAIA Agent/Chat/DT are software components, not new physical postulates. The central claim rests on untested reliability of the LLM agent and the RAG retrieval, not on fitted physics parameters.

free parameters (2)
  • RAG chunk size = 2048 tokens
    Section 2.1: documents are split into 2048-token chunks. This hand-chosen parameter directly affects retrieval quality and is not ablated.
  • Tool default parameters (STA/LTA windows, Geiger velocity model, Richter attenuation constants, ETAS parameters) = unspecified
    Section 2.2/3 rely on defaults of these algorithms; the agent's autonomous parameter selection is claimed but the exact values and their sensitivity are not reported.
axioms (3)
  • domain assumption Gemma3-27B-IT can reliably plan, tool-call, and complete multi-step geothermal workflows from natural-language prompts.
    Section 2.1 asserts the agent iteratively plans and executes steps; no error-rate or failure analysis is provided.
  • domain assumption The RAG corpus of ~5,000 geothermal papers covers the knowledge needed for correct answers.
    Section 2.1 describes the knowledge base but no retrieval precision/recall evaluation is reported.
  • domain assumption Wrapping standard tools (STA/LTA, Geiger, Richter, ETAS) through LLM-generated calls preserves their correctness.
    Sections 2.2 and 3 assume correct parameter passing and data formatting; no validation of the wrapper is reported.

pith-pipeline@v1.3.0-alltime-deepseek · 9516 in / 12867 out tokens · 113553 ms · 2026-08-03T23:49:24.052041+00:00 · methodology

0 comments
read the original abstract

Geothermal field development typically involves complex processes that require multi-disciplinary expertise in each process. Thus, decision-making often demands the integration of geological, geophysical, reservoir engineering, and operational data under tight time constraints. We present Geothermal Analytics and Intelligent Agent, or GAIA, an AI-based system for automation and assistance in geothermal field development. GAIA consists of three core components: GAIA Agent, GAIA Chat, and GAIA Digital Twin, or DT, which together constitute an agentic retrieval-augmented generation (RAG) workflow. Specifically, GAIA Agent, powered by a pre-trained large language model (LLM), designs and manages task pipelines by autonomously querying knowledge bases and orchestrating multi-step analyses. GAIA DT encapsulates classical and surrogate physics models, which, combined with built-in domain-specific subroutines and visualization tools, enable predictive modeling of geothermal systems. Lastly, GAIA Chat serves as a web-based interface for users, featuring a ChatGPT-like layout with additional functionalities such as interactive visualizations, parameter controls, and in-context document retrieval. To ensure GAIA's specialized capability for handling complex geothermal-related tasks, we curate a benchmark test set comprising various geothermal-related scenarios and rigorously evaluate the system's performance. Beyond task-level automation, GAIA is a unified framework that tightly couples physics-based modeling with agentic reasoning, which enables both domain-infused agentic evolutionary algorithm (meta-evolution) and end-to-end geothermal data processing within a single system. This integration highlights GAIA's potential not only as an assistive tool but as a platform for accelerating scientific discovery and enabling more autonomous, data-driven geothermal field development.

Figures

Figures reproduced from arXiv: 2511.03852 by Guodong Chen, Nori Nakata, Randy Harsuko, Zhengfa Bi.

Figure 1
Figure 1. Figure 1: GAIA Agentic RAG Workflow. Users interact with GAIA Chat interface by supplying prompts and [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: GAIA Chat interface. The main window features a chatbox to type and send messages, a conversation window [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Example of GAIA’s seismic waveform phase picking workflow. The user uploads a seismic waveform file and [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Example of GAIA’s event location estimation workflow. The user uploads a seismic waveform file and [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Example of GAIA’s event magnitude estimation workflow. The user uploads a seismic waveform file and [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Example of GAIA’s seismicity forecasting workflow. The user uploads a known seismicity catalog to the [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗

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

39 extracted references · 7 linked inside Pith

  1. [1]

    R. V. Allen. Automatic earthquake recognition and timing from single traces. Bulletin of the seismological society of America, 68 0 (5): 0 1521--1532, 1978

    1978

  2. [2]

    Beyreuther, R

    M. Beyreuther, R. Barsch, L. Krischer, T. Megies, Y. Behr, and J. Wassermann. Obspy: A python toolbox for seismology. Seismological Research Letters, 81 0 (3): 0 530--533, 2010

    2010

  3. [3]

    Z. Bi, N. Nakata, R. Nakata, P. Ren, X. Wu, and M. W. Mahoney. Advancing data-driven broadband seismic wavefield simulation with multi-conditional diffusion model. IEEE Transactions on Geoscience and Remote Sensing, 0 (99): 0 1--1, 2025

    2025

  4. [4]

    A. Chan, C. Ezell, M. Kaufmann, K. Wei, L. Hammond, H. Bradley, E. Bluemke, N. Rajkumar, D. Krueger, N. Kolt, et al. Visibility into ai agents. In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency, pages 958--973, 2024

    2024

  5. [5]

    Fomel, P

    S. Fomel, P. Sava, I. Vlad, Y. Liu, and V. Bashkardin. Madagascar: Open-source software project for multidimensional data analysis and reproducible computational experiments. Journal of Open Research Software, 1 0 (1): 0 e8--e8, 2013

    2013

  6. [6]

    L. Geiger. Probability method for the determination of earthquake epicentres from the arrival time only. Bull. St. Louis Univ., 8: 0 60, 1912

    1912

  7. [7]

    Gensheng, S

    L. Gensheng, S. Xianzhi, S. Yu, W. Gaosheng, et al. Current status and construction scheme of smart geothermal field technology. Petroleum Exploration and Development, 51 0 (4): 0 1035--1048, 2024

    2024

  8. [8]

    G \"u nther, S

    M. G \"u nther, S. Sturua, M. K. Akram, I. Mohr, A. Ungureanu, B. Wang, S. Eslami, S. Martens, M. Werk, N. Wang, et al. jina-embeddings-v4: Universal embeddings for multimodal multilingual retrieval. arXiv preprint arXiv:2506.18902, 2025

    Pith/arXiv arXiv 2025

  9. [9]

    Guti \'e rrez-Oribio, A

    D. Guti \'e rrez-Oribio, A. Stathas, and I. Stefanou. Ai-driven approach for sustainable extraction of earth's subsurface renewable energy while minimizing seismic activity. International Journal for Numerical and Analytical Methods in Geomechanics, 49 0 (4): 0 1126--1138, 2025

    2025

  10. [10]

    T. C. Hanks and H. Kanamori. A moment magnitude scale. Journal of Geophysical Research: Solid Earth, 84 0 (B5): 0 2348--2350, 1979

    1979

  11. [11]

    M. He, Q. Li, and X. Li. Injection-induced seismic risk management using machine learning methodology--a perspective study. Frontiers in Earth Science, 8: 0 227, 2020

    2020

  12. [12]

    Huang, F

    F. Huang, F. Wu, Z. Zhang, Q. Wang, L. Zhang, G. M. Boquet, and H. Chen. Geogpt. rag technical report. arXiv preprint arXiv:2509.09686, 2025 a

    arXiv 2025

  13. [13]

    Huang, Y

    Y. Huang, Y. Chen, H. Zhang, K. Li, M. Fang, L. Yang, X. Li, L. Shang, S. Xu, J. Hao, et al. Deep research agents: A systematic examination and roadmap. arXiv preprint arXiv:2506.18096, 2025 b

    Pith/arXiv arXiv 2025

  14. [14]

    Huang, T

    Y. Huang, T. Gao, H. Xu, Q. Zhao, Y. Song, Z. Gui, T. Lv, H. Chen, L. Cui, S. Li, et al. Peace: Empowering geologic map holistic understanding with mllms. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 3899--3908, 2025 c

    2025

  15. [15]

    Kanfar, A

    R. Kanfar, A. Alali, T.-L. Tonellot, H. Salim, and O. Ovcharenko. Intelligent seismic workflows: The power of generative ai and language models. The Leading Edge, 44 0 (2): 0 142--151, 2025

    2025

  16. [16]

    Khorasani, M

    M. Khorasani, M. Abdou, and J. H. Fern \'a ndez. Web application development with streamlit. Software Development, 498: 0 507, 2022

    2022

  17. [17]

    u ttler, M. Lewis, W.-t. Yih, T. Rockt \

    P. Lewis, E. Perez, A. Piktus, F. Petroni, V. Karpukhin, N. Goyal, H. K \"u ttler, M. Lewis, W.-t. Yih, T. Rockt \"a schel, et al. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in neural information processing systems, 33: 0 9459--9474, 2020

    2020

  18. [18]

    X. Li, S. Wang, S. Zeng, Y. Wu, and Y. Yang. A survey on llm-based multi-agent systems: workflow, infrastructure, and challenges. Vicinagearth, 1 0 (1): 0 9, 2024

    2024

  19. [19]

    Livathinos, C

    N. Livathinos, C. Auer, M. Lysak, A. Nassar, M. Dolfi, P. Vagenas, C. B. Ramis, M. Omenetti, K. Dinkla, Y. Kim, et al. Docling: An efficient open-source toolkit for ai-driven document conversion. arXiv preprint arXiv:2501.17887, 2025

    Pith/arXiv arXiv 2025

  20. [20]

    Lomax, J

    A. Lomax, J. Virieux, P. Volant, and C. Berge-Thierry. Probabilistic earthquake location in 3d and layered models: Introduction of a metropolis-gibbs method and comparison with linear locations. In Advances in seismic event location, pages 101--134. Springer, 2000

    2000

  21. [21]

    Lomax, A

    A. Lomax, A. Michelini, and A. Curtis. Earthquake location, direct, global-search methods. In Encyclopedia of complexity and systems science, pages 1--33. Springer, 2014

    2014

  22. [22]

    A. C. Montes, P. Ashok, and E. van Oort. Maximizing the value of geothermal data for reducing drilling risks

  23. [23]

    Nakata, S.-M

    N. Nakata, S.-M. Wu, C. Hopp, M. Robertson, and S. Dadi. Microseismicity observation and characterization at cape modern and utah forge. In Presented at the 49th Workshop on Geothermal Reservoir Engineering, 2024

    2024

  24. [24]

    Nakata, Z

    N. Nakata, Z. Bi, H. Qiu, C.-N. Liu, and R. Nakata. Ml-aided induced seismicity processing and interpretation for enhanced geothermal systems. The Leading Edge, 44 0 (4): 0 265--275, 2025

    2025

  25. [25]

    Y. Ogata. Statistical models for earthquake occurrences and residual analysis for point processes. Journal of the American Statistical association, 83 0 (401): 0 9--27, 1988

    1988

  26. [26]

    Introducing deep research (https://openai.com/index/introducing-deep-research/)

    OpenAI. Introducing deep research (https://openai.com/index/introducing-deep-research/). URL https://openai.com/index/introducing-deep-research/

  27. [27]

    B. Peng, C. Li, P. He, M. Galley, and J. Gao. Instruction tuning with gpt-4. arXiv preprint arXiv:2304.03277, 2023

    Pith/arXiv arXiv 2023

  28. [28]

    QGIS Geographic Information System

    QGIS Development Team . QGIS Geographic Information System. Open Source Geospatial Foundation, 2009. URL http://qgis.org

    2009

  29. [29]

    J. Qiu, X. Qi, T. Zhang, X. Juan, J. Guo, Y. Lu, Y. Wang, Z. Yao, Q. Ren, X. Jiang, et al. Alita: Generalist agent enabling scalable agentic reasoning with minimal predefinition and maximal self-evolution. arXiv preprint arXiv:2505.20286, 2025

    Pith/arXiv arXiv 2025

  30. [30]

    Reasenberg

    P. Reasenberg. Second-order moment of central california seismicity, 1969--1982. Journal of Geophysical Research: Solid Earth, 90 0 (B7): 0 5479--5495, 1985

    1969

  31. [31]

    C. F. Richter. An instrumental earthquake magnitude scale. Bulletin of the seismological society of America, 25 0 (1): 0 1--32, 1935

    1935

  32. [32]

    W. Shi, H. Tan, C. Kuang, X. Li, X. Ren, C. Zhang, H. Chen, Y. Wang, L. Shang, F. Yu, et al. Pangu deepdiver: Adaptive search intensity scaling via open-web reinforcement learning. arXiv preprint arXiv:2505.24332, 2025

    arXiv 2025

  33. [33]

    G. Team, P. Georgiev, V. I. Lei, R. Burnell, L. Bai, A. Gulati, G. Tanzer, D. Vincent, Z. Pan, S. Wang, et al. Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context. arXiv preprint arXiv:2403.05530, 2024

    Pith/arXiv arXiv 2024

  34. [34]

    G. Team, A. Kamath, J. Ferret, S. Pathak, N. Vieillard, R. Merhej, S. Perrin, T. Matejovicova, A. Ram \'e , M. Rivi \`e re, et al. Gemma 3 technical report. arXiv preprint arXiv:2503.19786, 2025

    Pith/arXiv arXiv 2025

  35. [35]

    S. A. Vaghefi, D. Stammbach, V. Muccione, J. Bingler, J. Ni, M. Kraus, S. Allen, C. Colesanti-Senni, T. Wekhof, T. Schimanski, et al. Chatclimate: Grounding conversational ai in climate science. Communications Earth & Environment, 4 0 (1): 0 480, 2023

    2023

  36. [36]

    Waldhauser

    F. Waldhauser. Hypodd-a program to compute double-difference hypocenter locations. Technical report, 2001

    2001

  37. [37]

    Weers, S

    J. Weers, S. Podgorny, N. Taverna, A. Anderson, S. Porse, and G. Buster. Empowering geothermal research: The geothermal data repository's new ai research assistant. Technical report, National Renewable Energy Laboratory (NREL), Golden, CO (United States), 2024

    2024

  38. [38]

    J. Wei, X. Wang, D. Schuurmans, M. Bosma, F. Xia, E. Chi, Q. V. Le, D. Zhou, et al. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35: 0 24824--24837, 2022

    2022

  39. [39]

    S. Yao, J. Zhao, D. Yu, N. Du, I. Shafran, K. Narasimhan, and Y. Cao. React: Synergizing reasoning and acting in language models. In International Conference on Learning Representations (ICLR), 2023

    2023