Pith. sign in

REVIEW 4 major objections 5 minor 35 references

Integrating Traditional Technical Analysis with AI: A Multi-Agent LLM-Based Approach to Stock Market Forecasting

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Integrating Elliott wave pattern recognition into a seven-agent LLM system enables faster and more accurate stock price forecasting, with DRL backtesting raising pattern validity by up to 16 percentage points.

desk verdict A well-described LLM-agent system for Elliott Wave analysis whose reported accuracy is built on a circular rule, so the empirical claims don't hold. read the letter →

arxiv 2506.16813 v1 pith:55GEZ5RY submitted 2025-06-20 cs.CE

Micha{l} Wawer , Jaros{l}aw A. Chudziak This is my paper
classification cs.CE
keywords Multi-AgentSystemsElliottWavePrincipleLargeLanguageModelsDeepReinforcementLearningStockMarketForecastingRetrieval-AugmentedGenerationTechnicalAnalysisPatternRecognition
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

This paper reports the design and evaluation of ElliottAgents, a multi-agent system in which large language models collaborate to apply the Elliott Wave Principle to stock market forecasting. The central claim is that this integration makes forecasting more accurate and interpretable than traditional methods alone, and that adding a deep-reinforcement-learning backtester improves pattern-recognition validity by up to 16 percentage points. The authors state that the experiments validate the system's effectiveness in pattern recognition and trend forecasting on historical data from major U.S. companies, and that the development has successfully addressed the primary research question. A sympathetic reader would take the contribution to be a demonstration that LLM-agent orchestration can operationalize a rule-based technical-analysis theory into actionable trading signals.

What carries the argument

The load-bearing machinery is the ElliottAgents architecture itself: a graph-orchestrated flow of LLM agents, each equipped with memory, planning, dynamic context, retrieval-augmented generation, and specialized tools. The Elliott Waves Analyst uses a dedicated pattern-recognition tool to detect non-overlapping impulsive waves; the Backtester uses Deep Reinforcement Learning trained on ten years of history to score candidate patterns; and the Investment Advisor applies Fibonacci proportions (approximately 1.62) to set price targets. The evaluation criteria are built from the same Elliott Wave rules the system forecasts with—a complete 1-2-3-4-5 wave is judged correct only if the following corrective wave A has exactly the same length and opposite direction of wave 5.

What would settle it

Run ElliottAgents' pattern detector on shuffled returns from the same stocks (or on synthetic random-walk prices) and compare the rate of detected 1-2-3-4-5 waves and the accuracy of its predicted wave-A lengths; if the detection rate and 'accuracy' do not drop materially relative to real data, the claimed predictive validity is an artifact of the evaluation rule rather than genuine forecast skill.

Watch

Extended reading notes

Core claim

The paper's core discovery is that a crew of seven specialized LLM agents—a coordinator, data engineer, Elliott waves analyst, backtester, technical analysis expert, investment advisor, and reports writer—can recognize Elliott impulse waves (1-2-3-4 and complete 1-2-3-4-5) in historical price data and turn them into forecasts of the next move. In cross-validation on 1,000 candlesticks per stock across daily and hourly intervals, the authors report that complete impulsive waves yield better predictions than incomplete ones, and that turning on the DRL backtester improves the fraction of predictions judged correct by up to 16 percentage points. The paper presents this as evidence that EWP can be integrated into a multi-agent architecture to more quickly and accurately predict future stock price movements, producing interpretable buy/sell signals with price targets.

Load-bearing premise

The reported accuracy scores depend on a definition of 'correct prediction' that is built from the same Elliott Wave rules the system uses to forecast, so the numbers measure how often the market conforms to the theory rather than whether the system beats an independent forecasting baseline.

Editorial extensions

If this is right

  • If the reported results hold, traders could use ElliottAgents to generate interpretable forecasts with concrete entry, exit, and stop-loss levels from LLM-agent analyses.
  • The reported improvement of up to 16 percentage points from DRL backtesting implies that continuous learning from historical patterns adds measurable predictive value beyond static pattern recognition.
  • The system's performance on daily and hourly intervals suggests the multi-agent approach extends Elliott Wave analysis across timeframes, though the authors note short-term accuracy is limited by noise and high-frequency trading.
  • Because the architecture's agents are modular, adding other technical-analysis methods or additional wave patterns would be straightforward, expanding the range of detectable market structures.
  • The paper's stated conclusion is that it successfully answered the research question of whether EWP can be integrated into a multi-agent LLM architecture to predict stock price movements more quickly and accurately.

Reading between the lines

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

  • A fair out-of-sample test would compare ElliottAgents' forecasts against a naive no-skill baseline, such as predicting that the previous trend continues, on held-out periods; the paper does not report such a comparison, so the Table 1 numbers should be read as theory-conformity rates rather than evidence of tradable alpha.
  • The same evaluation trap would apply to any extension of this architecture to other rule-based technical indicators: if 'correct' is defined by the theory's own rules, the metrics reward markets that happen to follow the theory rather than the system's skill.
  • The claimed continuous-learning benefit could be isolated by training the DRL backtester on the first half of the historical period and evaluating on the second half, testing whether the learning component generalizes or merely fits the training distribution.
  • The paper's own admission that short-term predictions are unreliable due to high-frequency trading noise suggests the truthful claim may be narrower: Elliott-wave LLM agents are a decision-support tool for swing traders, not a short-term forecasting engine.
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

4 major / 5 minor

Summary. The paper introduces ElliottAgents, a multi-agent LLM-based system that combines the Elliott Wave Principle (EWP) with RAG, DRL, and a graph-based knowledge store to identify impulsive and corrective wave patterns in stock data and generate trading recommendations. The authors report a case study on Apple and a cross-validation study on six U.S. stocks at daily and hourly intervals, with and without DRL backtesting. They conclude that the system 'successfully addressed the primary research question' and that DRL backtesting improves pattern recognition validity by up to 16 percentage points.

Significance. The architectural idea is timely: orchestrating specialized LLM agents for financial chart analysis, with structured memory and external tools, is a plausible direction for interpretable technical-analysis systems. The paper gives a reasonably concrete description of the agent roles, tools, and workflow, and it attempts to connect pattern recognition to backtesting. However, the quantitative evaluation is not adequate to support the central forecasting claim. The complete-wave success criterion is defined using the same Elliott Wave symmetry the system is designed to detect, so Table 1 measures label consistency rather than out-of-sample predictive skill. Sample sizes are small, baselines and statistical tests are absent, and the DRL training/evaluation split is unspecified. As it stands, the contribution is a system description with anecdotal evidence, not a validated forecasting method.

major comments (4)
  1. [Section 5] For complete 1-2-3-4-5 waves, a prediction is defined as accurate only if the subsequent corrective wave A 'exhibited a exact same length to that of wave 5' and moves in the opposite direction. This success criterion is exactly the Elliott Wave symmetry that the system's Elliott Waves Analyst is designed to enforce and detect. The hit rates in Table 1 therefore measure how often the future price path can be labeled as conforming to the EWP rule, not whether the system predicted future price movements in an independent sense. The criterion is also not operationalized: no tolerance is given for 'exact' equality, so it is unclear how any continuous price sequence could satisfy it or how the reported percentages were computed. An independent forecast target, such as the sign or size of future returns over a fixed horizon, together with a random or no-change baseline, is required for the central claim to be testable.
  2. [Section 5, Table 1] Even setting aside the circularity, the statistical evidence is insufficient. The number of detected patterns per stock ranges from 8 to 28, and several cells have fewer than 10 observations, so a single pattern changes the accuracy by more than 10 percentage points. No confidence intervals, significance tests, or effect sizes are reported, and there is no comparison against a baseline such as random direction, momentum, or a standard time-series model. Consequently, the discussion claim that backtesting improves pattern recognition validity 'by up to 16%' is not established; in several rows the improvement is zero or within sampling noise (e.g., hourly GOOG complete waves at 77.78% and hourly CSCO complete waves at 50.00% show no improvement with backtesting).
  3. [Section 5, second experimental phase] The DRL model is 'trained on 10 years of historical data for each company,' while the evaluation uses 1000 candlesticks per stock. The manuscript does not state whether the evaluation window is disjoint from the DRL training period or how the 1000 samples are split for cross-validation. If the evaluation period overlaps the training period, the reported accuracies are partly in-sample. The authors should specify the temporal split and, ideally, use a walk-forward or held-out period. Without this information, the backtesting results cannot be interpreted as evidence of out-of-sample predictive ability.
  4. [Abstract, Section 1, Section 7] The abstract and discussion state that the experimental results 'validate the system's effectiveness in pattern recognition and trend forecasting' and that the platform has 'successfully addressed the primary research question' of more quickly and accurately predicting future stock price movements. Given the circular evaluation metric, the small samples, and the lack of baselines, these statements are not supported by the reported data. The discussion also claims that the approach detects wave patterns 'with greater accuracy than comparable systems utilizing EWP (Tirea et al., 2012),' but no quantitative comparison with Tirea et al. or any other system is provided. The single AAPL case study is anecdotal and does not account for transaction costs, risk, or out-of-sample robustness.
minor comments (5)
  1. [Abstract] 'an multi-agent system' should be 'a multi-agent system'.
  2. [Section 5] 'exhibited a exact same length to that of wave 5' should be 'exhibited the exact same length as wave 5'.
  3. [Table 1] The percentage rounding is inconsistent: for example, 7/12 is 58.33%, not 58.34%, and 11/15 is 73.33%, not 73.34%.
  4. [Section 7] The paragraph beginning 'The development and testing of ElliottAgents have successfully addressed the primary research question posed at the outset of this study. The platform has demonstrated...' is repeated verbatim in two consecutive paragraphs.
  5. [Section 4.4 and Section 7] There are typos: 'Continous learning agent' should be 'Continuous learning agent', and 'backtesting capabilities usind DRL' should be 'using DRL'.

Circularity Check

1 steps flagged · score 7.0 of 10

The reported accuracy metric for complete 1-2-3-4-5 waves is circular: a prediction counts as correct only if the future forms a corrective wave A with exactly the same length as wave 5, an Elliott Wave rule rather than an independent price forecast target.

  1. self definitional [Section 5 (Experimental Setup and Methodology), evaluation criteria for complete impulse waves]
    "For complete waves (1-2-3-4-5), a prediction was considered accurate if the subsequent corrective wave A exhibited a exact same length to that of wave 5, moving in the opposite direction."

    The system detects 1-2-3-4-5 impulse patterns using the Elliott Wave Principle and generates predictions from that same framework. This evaluation rule defines 'correct' as the future price path containing a corrective wave A whose length exactly equals wave 5 and moves opposite, which is itself an Elliott Wave structural expectation. Thus the reported accuracies (50-89% in Table 1) measure how often subsequent price action can be labeled as an Elliott Wave A matching wave 5, not how well the system forecasts an independent quantity such as the sign or magnitude of future returns. The metric is built from the same wave-theoretic assumptions the system is supposed to validate, so the central prediction claim reduces to a self-consistent labeling check.

full rationale

Section 5 defines the correctness criterion for complete impulsive waves as the appearance of a subsequent corrective wave A with 'exact same length to that of wave 5' moving opposite. This is an Elliott Wave structural equality, not an independent forecast target such as realized return sign or magnitude; a future downward move of any other length would count as wrong. The system's wave detection and prediction are both built on the same EWP/Fibonacci framework, so the Table 1 accuracies measure the market's conformity to the labeling rule rather than out-of-sample forecasting skill. The incomplete-wave criterion (average direction over n = 1.62 first-wave lengths) is closer to a neutral target, but the horizon itself is also taken from the same Fibonacci/EWP relation and no random, no-change, or significance baseline is reported; the 'up to 16%' backtesting improvement is therefore an improvement on this theory-dependent metric. Self-citations such as (Chudziak and Wawer, 2024) are background and do not by themselves carry the forecasting claim. The central quantitative evidence of predictive accuracy reduces, for the complete-wave case, to a self-consistent EWP labeling check; hence the high circularity score.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central empirical claim rests on two untested domain assumptions: that the Elliott Wave Principle has predictive power, and that the detection tool is accurate. It also rests on two ad hoc evaluation rules (Fibonacci horizon and exact wave-length equality) that are not standard forecast metrics. The DRL component adds a training/evaluation setup whose data split is not disclosed. No free parameters are fitted to data in a conventional sense, but the evaluation itself is parameterized by theory-specific constants.

free parameters (2)
  • Fibonacci horizon multiplier = 1.62
    In Section 5, the prediction window n for incomplete 1-2-3-4 waves is set to approximately 1.62 times the length of the first wave. This constant is taken from Fibonacci/Elliott theory, not fitted to the data, but it directly determines which price bars are used to judge correctness.
  • Wave A length equality criterion = exact equality
    For complete 1-2-3-4-5 waves, a prediction is judged correct only if the next corrective wave A has exactly the same length as wave 5. This threshold is an arbitrary, theory-defined rule that defines success.
assumptions (3)
  • domain assumption Elliott Wave Principle is a meaningful description of market structure
    The entire system and evaluation assume that impulse and corrective wave patterns and Fibonacci length relationships have predictive content. This is contested in the finance literature and is not independently validated in the paper.
  • domain assumption The pattern-detection tool correctly identifies waves
    Section 5 says the system detects impulse and corrective patterns, but the accuracy of the detector is not measured against a labeled ground truth; the same detector's output is used as both the prediction and the evaluation target.
  • ad hoc to paper Exact length equality is a valid success criterion
    The rule that wave A must equal wave 5 in length is introduced specifically for this evaluation; it is not a standard forecasting metric like directional accuracy over a fixed horizon.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Integrating Traditional Technical Analysis with AI: A Multi-Agent LLM-Based Approach to Stock Market Forecasting." pith.science (2026). https://pith.science/paper/55GEZ5RY

@misc{pith2026250616813,
  author       = {Pith},
  title        = {Pith review of: Integrating Traditional Technical Analysis with AI: A Multi-Agent LLM-Based Approach to Stock Market Forecasting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/55GEZ5RY}},
  note         = {Machine review of arXiv:2506.16813}
}
read the original abstract

Traditional technical analysis methods face limitations in accurately predicting trends in today's complex financial markets. This paper introduces ElliottAgents, an multi-agent system that integrates the Elliott Wave Principle with AI for stock market forecasting. The inherent complexity of financial markets, characterized by non-linear dynamics, noise, and susceptibility to unpredictable external factors, poses significant challenges for accurate prediction. To address these challenges, the system employs LLMs to enhance natural language understanding and decision-making capabilities within a multi-agent framework. By leveraging technologies such as Retrieval-Augmented Generation (RAG) and Deep Reinforcement Learning (DRL), ElliottAgents performs continuous, multi-faceted analysis of market data to identify wave patterns and predict future price movements. The research explores the system's ability to process historical stock data, recognize Elliott wave patterns, and generate actionable insights for traders. Experimental results, conducted on historical data from major U.S. companies, validate the system's effectiveness in pattern recognition and trend forecasting across various time frames. This paper contributes to the field of AI-driven financial analysis by demonstrating how traditional technical analysis methods can be effectively combined with modern AI approaches to create more reliable and interpretable market prediction systems.

Figures

Figures reproduced from arXiv: 2506.16813 by the authors.

Figure 1
Figure 1. Basic trends in stock market, presented in (Mur [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. Fibonacci retracements in corrective waves (Frost [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Graph presenting data flow between agents, gen [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figures from the paper (5 more)
Figure 5
Figure 5. Figure 5: Overview of a LLM autonomous agent [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Asynchronous task execution enables agents to operate in parallel, mitigating bottlenecks and en￾hancing throughput (Li et al., 2024). Tasks such as backtesting and wave analysis, which do not re￾quire immediate interdependence, are executed con￾currently. This asynchr…
Figure 6
Figure 6. Figure 6: Interactions between agents and tools. leading to actionable strategies devised by the Invest￾ment Advisor and finalized in reports by the Reports Writer under the coordinator’s supervision. The designed flow is integral to its efficiency, scal￾ability, and resilience.…
Figure 7
Figure 7. Figure 7: Example interactions between agents analyzing [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Impulsive and corrective waves found on Apple [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

35 extracted references · 23 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION fin.entry add.period write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.state := if if FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 if FUNCTIO...

  2. [2]

    R., and Kumar, P

    Abrishami, S., Turek, M., Choudhury, A. R., and Kumar, P. (2019). Enhancing profit by predicting stock prices using deep neural networks. In 2019 IEEE 31st International Conference on Tools with Artificial Intelligence (ICTAI) , Portland, OR, USA

    work page 2019

  3. [3]

    and Oyetunji, O

    Akintola, K. and Oyetunji, O. (2021). Development of an agent-based framework for stock market trading. IRE Journals , 4(9)

    work page 2021

  4. [4]

    Auffarth, B. (2023). Generative AI with LangChain, Build large language model (LLM) apps with Python, ChatGPT, and other LLMs . Packt Publishing

    work page 2023

  5. [5]

    Boroden, C. (2008). Fibonacci Trading: How to Master the Time and Price Advantage . McGraw Hill

    work page 2008

  6. [6]

    Chen, H. (2024). Understand the llm agent orchestration. https://medium.com/scisharp/understand-the-llm-agent-orchestration-043ebfaead1f Accessed: Jun. 1, 2024

    work page 2024

  7. [7]

    Cheng, Y., Zhang, C., Zhang, Z., Meng, X., Hong, S., Li, W., Wang, Z., Wang, Z., Yin, F., Zhao, J., and He, X. (2024). Exploring large language model based intelligent agents: Definitions, methods, and prospects. arXiv:2401.03428v1 [cs.AI]

    arXiv 2024

  8. [8]

    Chi-Min Chan, Weize Chen, Y. S. J. Y. W. X. S. Z. J. F. Z. L. (2023). Chateval: Towards better llm-based evaluators through multi-agent debate. arXiv:2308.07201v1 [cs.CL]

    arXiv 2023

Show all 35 references
  1. [9]

    Chudziak, A. (2023). Predictability of stock returns using neural networks: Elusive in the long term. Expert Systems with Applications , 213

    2023

  2. [10]

    Chudziak, J. A. and Wawer, M. (2024). Elliottagents: A natural language-driven multi-agent system for stock market analysis and prediction. In Proceedings of the 38th Pacific Asia Conference on Language, Information and Computation , Tokyo, Japan, (in press)

    2024

  3. [11]

    J., Jr., R

    Frost, A. J., Jr., R. R. P., and Collins, C. J. (2001). Elliott Wave Principle: Key to Market Behavior . Wiley

    2001

  4. [12]

    A., El-fouly, R

    Gamil, A. A., El-fouly, R. S., and Darwish, N. M. (2007). Stock technical analysis using multi agent and fuzzy logic. In Proceedings of the World Congress on Engineering, WCE 2007 , London, UK

    2007

  5. [13]

    Guo, T. et al. (2024). Large language model based multi-agents: A survey of progress and challenges. arXiv:2402.01680v2 [cs.CL]

    2024 arXiv

  6. [14]

    Hu, B., Zhao, C., Zhang, P., Zhou, Z., Yang, Y., Xu, Z., and Liu, B. (2024). Enabling intelligent interactions between an agent and an llm: A reinforcement learning approach. arXiv:2306.03604v8 [cs.AI]

    2024 arXiv

  7. [15]

    Jin, M., Tang, H., Zhang, C., Yu, Q., Liu, C., Zhu, S., Zhang, Y., and Du, M. (2024). Time series forecasting with llms: Understanding and enhancing model capabilities. arXiv:2402.10835v2 [cs.CL]

    2024 arXiv

  8. [16]

    and Duman, E

    Kabbani, T. and Duman, E. (2022). Deep reinforcement learning approach for trading automation in the stock market. IEEE Access , 10

    2022

  9. [17]

    Lapan, M. (2020). Deep Reinforcement Learning Hands-On Second Edition . Packt

    2020

  10. [18]

    and Truitt, S

    Larson, J. and Truitt, S. (2024). Graphrag: Unlocking llm discovery on narrative private data. https://www.microsoft.com/en-us/research/blog/graphrag-unlocking-llm-discovery-on-narrative-private-data Accessed: May. 10, 2024

    2024

  11. [19]

    Lewis, P. et al. (2021). Retrieval-augmented generation for knowledge-intensive nlp tasks. arXiv:2005.11401v4 [cs.CL]

    2021 arXiv

  12. [20]

    Li, J., Zhang, Q., Yu, Y., Fu, Q., and Ye, D. (2024). More agents is all you need. arXiv:2402.05120v1 [cs.CL]

    2024 arXiv

  13. [21]

    Luo, Y., Liu, K., and Davis, D. N. (2002). A multi-agent decision support system for stock trading. IEEE Network , 16(1)

    2002

  14. [22]

    Lussange, J., Lazarevich, I., Bourgeois-Gironde, S., Palminteri, S., and Gutkin, B. (2020). Modelling stock markets by multi-agent reinforcement learning. Computational Economics . hal-03055070

    2020

  15. [23]

    Minsky, M. (1988). The Society of Mind . Simon & Schuster

    1988

  16. [24]

    Murphy, J. J. (1999). Technical Analysis of the Financial Markets: A Comprehensive Guide to Trading Methods and Applications . New York Institute of Finance

    1999

  17. [25]

    and Norvig, P

    Russell, S. and Norvig, P. (1995). Artificial Intelligence: A Modern Approach . Prentice Hall

    1995

  18. [26]

    Szydlowski, K. L. and Chudziak, J. A. (2024). Toward predictive stock trading with hidformer integrated into reinforcement learning strategy. In Proceedings of the 36th International Conference on Tools for Artificial Intelligence (ICTAI 2024) , Herndon, VA, USA, (in press)

    2024

  19. [27]

    A., Gupta, V., Althoff, T., and Hartvigsen, T

    Tan, M., Merrill, M. A., Gupta, V., Althoff, T., and Hartvigsen, T. (2024). Are language models actually useful for time series forecasting? arXiv:2406.16964v1 [cs.LG]

    2024 arXiv

  20. [28]

    Thomas, R. J. (2024). The rise of large action models, lams: How ai can understand and execute human intentions? https://medium.com/version-1/the-rise-of-large-action-models-lams-how-ai-can-understand-and-execute-human-intentions-f59c8e78bc09 Accessed: Jun. 20, 2024

    2024

  21. [29]

    Tirea, M., Tandau, I., and Negru, V. (2012). Stock market multi-agent recommendation system based on the elliott wave principle. In International Conference on Availability, Reliability, and Security , Prague, Czech Republic

    2012

  22. [30]

    Tsay, R. S. (2010). Analysis of Financial Time Series Third Edition . Wiley

    2010

  23. [31]

    Wang, L., Zhang, X., Su, H., and Zhu, J. (2024). A comprehensive survey of continual learning: Theory, method and application. arXiv:2302.00487 [cs.LG]

    2024 arXiv

  24. [32]

    Weng, L. (2023). Llm-powered autonomous agents. lilianweng.github.io

    2023

  25. [33]

    Wittkampf, F. (2024). Next-level agents: Unlocking the power of dynamic context. https://towardsdatascience.com/next-level-agents-unlocking-the-power-of-dynamic-context-68b8647eef89 Accessed: Jun. 1, 2024

    2024

  26. [34]

    Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K., and Cao, Y. (2023). React: Synergizing reasoning and acting in language models. arXiv:2210.03629 [cs.CL]

    2023 arXiv

  27. [35]

    Zhao, P., Jin, Z., and Cheng, N. (2023). An in-depth survey of large language model-based artificial intelligence agents. arXiv:2309.14365v1 [cs.CL]

    2023 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.