Pith. sign in

REVIEW 4 major objections 5 minor 33 references

A Multimodal Social Agent

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read MuSA, a modular LLM-based agent, improves social content analysis by adding optimize-criticize-refine cycles to a reasoner-actor baseline.

desk verdict A clean modular integration of known LLM-agent techniques for social content analysis, but the paper's own tables show the gains come from a stronger optimizer model, not from MuSA's architecture. read the letter →

arxiv 2501.06189 v1 pith:OCEN6SIP submitted 2024-12-11 cs.AI cs.CL

Athina Bikaki , Ioannis A. Kakadiaris This is my paper · ORCID
classification cs.AIcs.CL
keywords LLMagentsmultimodalsocialreasoningverbalreinforcementquestionansweringtitlegenerationcontentcategorizationchain-of-thought
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 claims that a modular LLM-based agent, MuSA, improves analysis of online social content by separating reasoning, planning, optimization, critique, and refinement into independent units instead of relying on a single reason-then-answer call. The authors evaluate MuSA on question answering, title generation, and content categorization and report that the full pipeline with optimize-criticize-refine cycles performs substantially better than a reasoner-actor baseline. The significance is that this is a model-agnostic, extensible design that lets practitioners combine weak, cheap models for acting and strong models for optimization, and it points toward agents that can be assembled for other human-centric content tasks without fine-tuning.

What carries the argument

The load-bearing machinery is a plan-then-optimize loop. A planner builds a sequence of actions; a text-based optimizer, operating like gradient descent over natural language, rewrites the plan or answer; and if the Jensen-Shannon divergence between the planner's and optimizer's responses exceeds a threshold, a critic chooses the better response and a refiner converts the feedback into instructions for the next planning round. This cycle happens before and during action execution, so the final answer is the actor's second attempt after optimization.

What would settle it

If an independent replication on larger, randomly drawn subsets of the same three datasets, with confidence intervals, found that the full MuSA pipeline does not beat a reasoner-actor baseline on exact match, BLEU-4, or accuracy, the paper's central claim would be disproved.

Watch

Extended reading notes

Core claim

MuSA's central claim is that adding an optimize-criticize-refine cycle to a reasoner-actor pipeline yields better answers, titles, and category labels for text-rich social content. In the reported experiments, exact match on HotpotQA rises from 60 to 62, BLEU-4 on WikiWeb2M rises from 39.9 to 41.6, and accuracy on MN-DS level-1 categorization rises from 48.1 to 50.8. The authors interpret this as evidence that complex social content benefits from explicit plan generation, textual optimization of the response, and a divergence-gated critic/refiner that intervenes only when the original and optimized responses disagree.

Load-bearing premise

The paper's results depend on the assumption that its hand-picked evaluation subsets — 100 HotpotQA questions, 100 WikiWeb2M pages, and 218 MN-DS articles — fairly represent the social-content tasks, even though the selection protocol is not specified and no confidence intervals are reported.

Editorial extensions

If this is right

  • Modular units let developers pair cheap open models for acting with stronger models for optimization, lowering cost without losing accuracy.
  • The divergence gate means the critic and refiner run only when the planner and optimizer disagree, saving computation on straightforward cases.
  • Adding visual input improves title generation, and using a stronger multimodal model as the optimizer yields further gains.
  • Reasoning strategy must be chosen per model: chain-of-thought with self-reflection helps Gemini-1.5-Flash-8B but hurts Mistral-based models in the reported QA results.

Reading between the lines

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

  • The threshold on Jensen-Shannon divergence is effectively a hyperparameter; a natural extension is to tune it per task and measure the accuracy-versus-compute tradeoff.
  • Because the evaluation subsets are small, unreleased, and single-draw, an independent multi-seed replication with confidence intervals is needed to tell whether the observed gains are general.
  • The same architecture could be extended to other social-content actions, such as stance detection, misinformation flagging, or social-media thread summarization, where the critic/refiner loop could be applied to domain-specific outputs.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 MuSA, a multimodal LLM-based agent whose computing engine combines reason, plan, optimize, criticize, refine, and act units for social content analysis tasks. The authors evaluate MuSA on question answering (HotpotQA), title generation (WikiWeb2M), and content categorization (MN-DS), reporting that the full pipeline performs better than a reason-then-act baseline. The central claim is that the modular combination of these units substantially improves performance over the authors' own baselines.

Significance. If the reported effects are genuine, the paper would demonstrate a useful modular agent architecture that separates planner, optimizer, critic, and refiner roles and applies TextGrad-based prompt/plan optimization to social content analysis. The paper has several strengths: Algorithm 1 gives a concrete specification of the proposed loop, the evaluation uses public datasets, and the ablations of reasoning strategies (Section 4.2.1) are informative. However, the empirical support as presented is too weak to establish the central claim. The apparent gains are confounded by the identity of the optimizer model, the critic/refiner cycle—a distinctive part of the architecture—is never evaluated, no external baselines are provided, and the small hand-selected subsets come without significance tests or confidence intervals. The contribution therefore remains a plausible design proposal rather than a demonstrated improvement.

major comments (4)
  1. [§4.2.2 (Table 6) and §4.2.4 (Table 8)] The reported improvements of the A→O→A pipeline over A alone are confounded by model identity, not architecture. In Table 6, the largest EM gain (60→64) occurs when the optimizer is the stronger Gemini-Pro, whereas with the same Gemini-1.5-Flash-8B as both actor and optimizer, EM rises only from 60 to 62 and F1 drops from 72.1 to 67.6. In Table 8, with Gemini-1.5-Flash-8B in both roles, BLEU-4 falls from 39.9 to 35.2 and ROUGE-L F1 from 60.4 to 51.3 relative to the actor alone, while the gains in Table 3 are obtained with Pixtral-12B as the optimizer. These results are consistent with the simpler explanation that putting a stronger LLM in the loop helps, rather than with MuSA's modular reason/plan/criticize/refine design. The primary evidence should be same-model comparisons, and the optimizer's contribution should be evaluated at matched model capability.
  2. [§3.4, §3.5, Algorithm 1] The critic and refiner units, which are central to the claimed MuSA architecture, are never isolated in any experiment. The only pipeline variant tested is A→O→A (Tables 2, 3, 4, 6, 8); no experiment activates the CRITICIZE and REFINE steps of Algorithm 1. Consequently, the claim that criticizing and refining improve performance is unsupported. The threshold θ that controls critic activation is also never specified or varied. The authors should either provide experiments that turn the critic/refiner cycle on and off, or explicitly restrict the empirical claims to the reason–optimize–act pipeline.
  3. [§4.1.2, §4.1.4, §4.1.5] The evaluation subsets are hand-selected and not released, with no concrete selection criteria beyond 'closely related to the social domain' (HotpotQA) and 'randomly selected' (WikiWeb2M). No confidence intervals, error bars, or significance tests are reported for any metric. Given the small sample sizes (100, 100, and 218 examples) and the small absolute differences (e.g., EM 60→62 in Table 2), the observed gains may not be statistically reliable. The authors should specify the selection protocol, release the subsets, and report variance or significance.
  4. [§4] All comparisons are against the authors' own ablations; no external baselines are provided. The abstract's claim that 'MuSA performs substantially better than our baselines' is therefore a self-referential comparison and does not position MuSA relative to existing methods. At a minimum, the paper should compare against standard approaches on these datasets (e.g., ReAct, Reflexion, or TextGrad applied to the actor alone) or cite previously published numbers on the same subsets.
minor comments (5)
  1. [Abstract] The abstract contains a duplicated sentence: 'However, the potential of integrating computers with these social capabilities is still relatively unexplored.' appears twice in succession.
  2. [§3.4] The sentence 'In cases where the similarity of responses falls below the threshold θ, a critic will not be called' is ambiguous or contradictory: JSD is a divergence (higher values mean more dissimilar), and Algorithm 1 calls the critic when JSD ≥ θ. The wording should be corrected to avoid confusion.
  3. [Tables 3 and 8] The column heading 'F1 RL P RL R' is unclear; the table should explicitly name the ROUGE-L submetrics (e.g., ROUGE-L F1, ROUGE-L Precision, ROUGE-L Recall).
  4. [Related Work] Reference [7] is cited for CRITIC in Section 2, but the CRITIC reference is [6] (Gou et al.); the in-text citation should be corrected.
  5. [§3.7] The text says external tools are used in specific actions, but Table 1 marks ET as '×' for all actions and no experiment uses external tools. This should be clarified as a design option or future work.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: MuSA's evaluation is self-contained against external datasets and its own ablations; no fitted constants or self-citations feed the central claim.

full rationale

The paper's central comparative claim is that the MuSA pipeline with reason, plan, optimize, criticize, refine, and act units performs better than simpler baselines. This claim is tested on external, human-annotated datasets (HotpotQA, WikiWeb2M, MN-DS), and the baselines are the authors' own ablations, which is a standard within-paper experimental comparison rather than a derivation that reduces to its inputs. No parameter is fitted to the target metric and then reported as a prediction: the JSD threshold theta is a design choice, not calibrated on held-out labels, and no uniqueness theorem or self-citation is invoked to force the architecture. The paper cites external prior work (TextGrad, Self-Refine, Reflexion, CoT) as building blocks, but none of those citations is by the present authors, and the cited results are independently published rather than self-supporting. Concerns about subset representativeness and the confound that gains appear mainly when a stronger model serves as the optimizer are empirical validity issues, not circularity. Under the stated criteria, no load-bearing step is equivalent to its input by construction, so the appropriate finding is no significant circularity.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

No new physical or mathematical entities are introduced. MuSA is a software system built from existing components, not a postulate requiring independent evidence.

free parameters (2)
  • Critic activation threshold theta = not reported
    Used in Algorithm 1 and Section 3.4 to decide whether the critic and refiner run; hand-chosen and never reported, directly controls how often the optimize-criticize-refine loop fires.
  • TextGrad optimization hyperparameters (learning rate alpha, number of iterations) = not reported
    Appear in Equations 2 and 3 and Section 3.3; values are not given, yet they determine how much the optimizer changes plans and answers.
assumptions (4)
  • domain assumption The environment is closed and static.
    Section 3 states 'we assume a closed static environment E'; real social media is open and dynamic, so the planning formalization may not transfer.
  • domain assumption TextGrad's text-based gradients improve LLM outputs for these tasks.
    Section 3.3 imports TextGrad's optimization as a reliable component; the paper does not independently validate this assumption on social content.
  • ad hoc to paper Jensen-Shannon divergence between planner and optimizer responses is a valid proxy for when critique is needed.
    Section 3.4 uses JSD to gate the critic; no evidence is given that JSD correlates with output quality, and the threshold is unreported.
  • ad hoc to paper The hand-selected test subsets are representative of the tasks.
    Section 4.1 selects 100 HotpotQA, 100 WikiWeb2M, and 218 MN-DS examples with vague criteria; no subsets are released, so representativeness is assumed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Multimodal Social Agent." pith.science (2026). https://pith.science/paper/OCEN6SIP

@misc{pith2026250106189,
  author       = {Pith},
  title        = {Pith review of: A Multimodal Social Agent},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OCEN6SIP}},
  note         = {Machine review of arXiv:2501.06189}
}
read the original abstract

In recent years, large language models (LLMs) have demonstrated remarkable progress in common-sense reasoning tasks. This ability is fundamental to understanding social dynamics, interactions, and communication. However, the potential of integrating computers with these social capabilities is still relatively unexplored. However, the potential of integrating computers with these social capabilities is still relatively unexplored. This paper introduces MuSA, a multimodal LLM-based agent that analyzes text-rich social content tailored to address selected human-centric content analysis tasks, such as question answering, visual question answering, title generation, and categorization. It uses planning, reasoning, acting, optimizing, criticizing, and refining strategies to complete a task. Our approach demonstrates that MuSA can automate and improve social content analysis, helping decision-making processes across various applications. We have evaluated our agent's capabilities in question answering, title generation, and content categorization tasks. MuSA performs substantially better than our baselines.

Figures

Figures reproduced from arXiv: 2501.06189 by the authors.

Figure 1
Figure 1. An example for MuSA. Post was selected from X [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Visualization of MuSA plan and action execution for our selected content analysis tasks (T). The similarity between [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 5
Figure 5. An example of a multiple-choice question from a [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Decision plan of a task. Plan A was proposed by the [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 7
Figure 7. Figure 7: Title generation using an example from Wiki [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: Most discrepant Level-1 (L) and Level-2 (R) categories from our selection of articles in the MN-DS [ [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 23 canonical work pages

  1. [1]

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey Wu, Clemens Winter, Chris Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott Gr...

    work page 2020

  2. [2]

    Andrea Burns, Krishna Srinivasan, Joshua Ainslie, Geoff Brown, Bryan Plummer, Kate Saenko, Jianmo Ni, and Mandy Guo. 2023. A Suite of Generative Tasks for Multi-Level Multimodal Webpage Understanding. In Proc. 2023 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, Singapore, 1917–1947. https://doi.o...

    doi:10.18653/v1/2023.emnlp- 2023

  3. [3]

    CDC. 2023. Principles of Epidemiology: Lesson 5 Quiz|Self-Study Course SS1978|CDC. CDC. Retrieved 2024-10-27 from https://archive.cdc.gov/www_ cdc_gov/csels/dsepd/ss1978/lesson5/quiz.html

    work page 2023

  4. [4]

    Yingqian Cui, Pengfei He, Xianfeng Tang, Qi He, Chen Luo, Jiliang Tang, and Yue Xing. 2024. A Theoretical Understanding of Chain-of-Thought: Coherent Reasoning and Error-Aware Demonstration. https://doi.org/10.48550/arXiv. 2410.16540

    arXiv doi:10.48550/arxiv.2410.16540 2024

  5. [5]

    Google DeepMind. 2024. Gemini Models. Google. Retrieved 2024-11-26 from https://deepmind.google/technologies/gemini/

    work page 2024

  6. [6]

    Zhibin Gou, Zhihong Shao, Yeyun Gong, yelong shen, Yujiu Yang, Nan Duan, and Weizhu Chen. 2024. CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing. In Proc. 12th International Conference on Learning Representations. OpenReview.net, Vienna, Austria, 9 pages. https://openreview. net/forum?id=Sx038qxjek

    work page 2024

  7. [7]

    Qingyan Guo, Rui Wang, Junliang Guo, Bei Li, Kaitao Song, Xu Tan, Guoqing Liu, Jiang Bian, and Yujiu Yang. 2024. Connecting Large Language Models with Evolutionary Algorithms Yields Powerful Prompt Optimizers. In Proc. 12th International Conference on Learning Representations . OpenReview.net, Vienna, Austria, 9 pages. https://openreview.net/forum?id=ZG3RaNIsO8

    work page 2024

  8. [8]

    Xu Huang, Weiwen Liu, Xiaolong Chen, Xingmei Wang, Hao Wang, Defu Lian, Yasheng Wang, Ruiming Tang, and Enhong Chen. 2024. Understanding the planning of LLM agents: A survey. https://doi.org/10.48550/arXiv.2402.02716 arXiv:2402.02716

    arXiv doi:10.48550/arxiv.2402.02716 2024

Show all 33 references
  1. [9]

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

    Omar Khattab, Arnav Singhvi, Paridhi Maheshwari, Zhiyuan Zhang, Keshav Santhanam, Sri Vardhamanan A, Saiful Haq, Ashutosh Sharma, Thomas T. Joshi, Hanna Moazam, Heather Miller, Matei Zaharia, and Christopher Potts

  2. [10]

    Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. 2022. Large language models are zero-shot reasoners. In Proc. of the 36th International Conference on Neural Information Processing Systems . Curran Associates Inc., New York, USA, 22199–22213

    2022

  3. [11]

    Yunshi Lan, Xiang Li, Xin Liu, Yang Li, Wei Qin, and Weining Qian. 2023. Im- proving Zero-shot Visual Question Answering via Large Language Models with Reasoning Question Prompts. In Proc. 31st ACM International Conference on Multimedia. Association for Computing Machinery, Ne...

    2023

  4. [12]

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2020. Retrieval-augmented gen- eration for knowledge-intensive NLP tasks. In Proc. 34th...

    2020

  5. [13]

    J. Lin. 1991. Divergence measures based on the Shannon entropy. IEEE Trans- actions on Information Theory 37, 1 (1991), 145–151. https://doi.org/10.1109/18. 61115

    1991 doi

  6. [14]

    Peter Lindes and John E. Laird. 2016. Toward integrating cognitive linguistics and cognitive language processing. In Proc. 14th International Conference on Cognitive Modeling. Penn State, Pennsylvania, USA, 86–92. http://acs.ist.psu. edu/iccm2016/proceedings/lindes2016iccm.pdf

    2016

  7. [15]

    Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, Shashank Gupta, Bodhisattwa Prasad Majumder, Katherine Hermann, Sean Welleck, Amir Yazdanbakhsh, and Peter Clark. 2023. Self-Refine: It...

    2023

  8. [16]

    Mistral. 2024. Mistral AI Large Language Models . Mistral. Retrieved 2024-11-26 from https://docs.mistral.ai/getting-started/models/models_overview/

    2024

  9. [17]

    2004.Automated Planning: Theory & Practice

    Dana Nau, Malik Ghallab, and Paolo Traverso. 2004.Automated Planning: Theory & Practice. Morgan Kaufmann Publishers Inc., San Francisco, CA, USA

    2004

  10. [18]

    Alina Petukhova and Nuno Fachada. 2023. MN-DS: A Multilabeled News Dataset for News Articles Hierarchical Classification. Data 8, 5 (2023), 74. https://doi. org/10.3390/data8050074

    2023 doi

  11. [19]

    Jianing Qiu, Kyle Lam, Guohao Li, Amish Acharya, Tien Yin Wong, Ara Darzi, Wu Yuan, and Eric J. Topol. 2024. LLM-based agentic systems in medicine and A Multimodal Social Agent healthcare. Nature Machine Intelligence 6 (2024), 1–3. https://doi.org/10.1038/ s42256-024-00944-1

    2024

  12. [20]

    Hao Shao, Shengju Qian, Han Xiao, Guanglu Song, Zhuofan Zong, Letian Wang, Yu Liu, and Hongsheng Li. 2024. Visual CoT: Advancing Multi-Modal Language Models with a Comprehensive Dataset and Benchmark for Chain-of-Thought Reasoning. https://doi.org/10.48550/arXiv.2403.16999 arX...

    2024 doi arXiv

  13. [21]

    Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. Reflexion: Language Agents with Verbal Reinforcement Learning. In Proc. 36th International Conference on Neural Information Processing Systems, Vol. 36. Curran Associates, Inc., New Orlean...

    2023

  14. [22]

    Le, and Denny Zhou

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc V. Le, and Denny Zhou. 2022. Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. In Proc. 35th International Conference on Neural Information Processing Systems , Vol...

    2022

  15. [23]

    World Health Organization (WHO). 2023. Sun exposure and non-melanoma skin cancer. WHO. Retrieved 2024-10-26 from https://www.facebook.com/WHO

    2023

  16. [24]

    2024.Delegates at the World Health Assembly #WHA77 approved a four-year

    World Health Organization (WHO). 2024.Delegates at the World Health Assembly #WHA77 approved a four-year... WHO. Retrieved 2024-10-25 from https://x.com/ WHO/status/1795554378511773795

    2024

  17. [25]

    Le, Denny Zhou, and Xinyun Chen

    Chengrun Yang, Xuezhi Wang, Yifeng Lu, Hanxiao Liu, Quoc V. Le, Denny Zhou, and Xinyun Chen. 2024. Large Language Models as Optimizers. In Proc. 12th International Conference on Learning Representations . OpenReview.net, Vienna, Austria, 9 pages. https://openreview.net/forum?i...

    2024

  18. [26]

    Jingfeng Yang, Hongye Jin, Ruixiang Tang, Xiaotian Han, Qizhang Feng, Haoming Jiang, Shaochen Zhong, Bing Yin, and Xia Hu. 2024. Harnessing the Power of LLMs in Practice: A Survey on ChatGPT and Beyond. ACM Transactions on Knowledge Discovery from Data 18, 6 (2024), 160:1–160:...

    2024

  19. [27]

    Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, and Christopher D. Manning. 2018. HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering. In Proc. 2018 Conference on Empirical Methods in Natural Language Processing. ...

    2018 doi

  20. [28]

    Narasimhan, and Yuan Cao

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik R. Narasimhan, and Yuan Cao. 2023. ReAct: Synergizing Reasoning and Acting in Language Models. In Proc. 11th International Conference on Learning Representa- tions. OpenReview.net, Kigali, Rwanda, 9 pages. https...

    2023

  21. [29]

    Differentiation

    Mert Yuksekgonul, Federico Bianchi, Joseph Boen, Sheng Liu, Zhi Huang, Carlos Guestrin, and James Zou. 2024. TextGrad: Automatic "Differentiation" via Text. https://doi.org/10.48550/arXiv.2406.07496

    2024 doi arXiv

  22. [30]

    Danyang Zhang, Lu Chen, Situo Zhang, Hongshen Xu, Zihan Zhao, and Kai Yu

  23. [31]

    Wanjun Zhong, Lianghong Guo, Qiqi Gao, He Ye, and Yanlin Wang. 2024. Mem- oryBank: Enhancing Large Language Models with Long-Term Memory. In Proc. 38th AAAI Conference on Artificial Intelligence , Vol. 38. AAAI Press, Vancouver, Canada, 19724–19731. https://doi.org/10.1609/aaa...

    2024 doi

  24. [2023]

    Large Language Models Are Semi-Parametric Reinforcement Learning Agents. In Proc. 37th International Conference on Neural Information Processing Systems. Curran Associates, Inc., New Orleans, USA, 9 pages. https://openreview. net/forum?id=ZcJa1R6j3v

  25. [2024]

    DSPy: Compiling Declarative Language Model Calls into State-of-the- Art Pipelines. In Proc. 12th International Conference on Learning Representations . OpenReview.net, Vienna, Austria, 9 pages. https://openreview.net/forum?id= sY5N0zY5Od

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.