pith. machine review for the scientific record. sign in

arxiv: 2605.03308 · v1 · submitted 2026-05-05 · 💻 cs.AI

Recognition: unknown

Revisiting the Travel Planning Capabilities of Large Language Models

Bo-Wen Zhang , Jin Ye , Peng-Yu Hua , Jia-Wei Cao , Jie-Jing Shao , Yu-Feng Li , Lan-Zhe Guo
Authors on Pith no claims yet

Pith reviewed 2026-05-07 16:48 UTC · model grok-4.3

classification 💻 cs.AI
keywords large language modelstravel planningreasoning capabilitiesconstraint extractionplan generationself-correctionbenchmark evaluationimplicit requirements
0
0 comments X

The pith

Large language models extract explicit travel constraints accurately but fail to infer implicit requirements and correct their own planning errors effectively.

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

The paper decomposes travel planning into five separate skills—Constraint Extraction, Tool Use, Plan Generation, Error Identification, and Error Correction—to test LLMs in isolation rather than judging only the final itinerary. Using oracle-provided intermediate steps, it measures each skill without the usual cascade of earlier mistakes. Results show strong performance on pulling out stated rules but clear weakness at deducing unstated, real-world needs that any human traveler would consider. Plans also display consistent structural biases, and attempts at self-correction tend to overreact or cling to flawed choices. Pinpointing these separate weaknesses gives concrete targets for improving long-horizon reasoning in LLMs.

Core claim

By decomposing travel planning into the atomic sub-capabilities of Constraint Extraction, Tool Use, Plan Generation, Error Identification, and Error Correction, and evaluating each one separately with oracle intermediate contexts, the study establishes that LLMs are proficient at extracting explicit constraints but struggle to infer implicit open-world requirements, exhibit structural biases during plan generation, and perform ineffective self-correction marked by excessive sensitivity and erroneous persistence.

What carries the argument

The five atomic sub-capabilities and the decoupled evaluation protocol that supplies oracle intermediate contexts to isolate each capability without cascading errors.

If this is right

  • Training methods must target implicit requirement inference separately from explicit constraint handling.
  • Plan generation modules need mechanisms to counteract structural biases that appear even in isolation.
  • Self-correction loops require redesign to reduce over-sensitivity and error persistence.
  • Future benchmarks should adopt decoupled protocols to diagnose specific failure modes rather than relying on end-to-end plan quality.
  • Improvements on these atomic skills would directly raise performance on other long-horizon reasoning tasks.

Where Pith is reading between the lines

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

  • Real-world deployment of LLM planners would still require human review for unspoken constraints such as weather or personal preferences.
  • The same decomposition approach could diagnose weaknesses in LLM performance on project scheduling or scientific experiment design.
  • Models trained with explicit signals for open-world inference might close the gap observed here.
  • User studies with actual travelers could test whether the isolated weaknesses produce plans that fail in practice.

Load-bearing premise

The five chosen sub-capabilities fully cover travel planning and the oracle contexts isolate each skill without introducing new biases or artificial advantages.

What would settle it

An LLM that achieves comparable accuracy on inferring implicit open-world requirements as on explicit constraints when tested in the same isolated, oracle-provided setup would falsify the reported performance contrast.

Figures

Figures reproduced from arXiv: 2605.03308 by Bo-Wen Zhang, Jia-Wei Cao, Jie-Jing Shao, Jin Ye, Lan-Zhe Guo, Peng-Yu Hua, Yu-Feng Li.

Figure 1
Figure 1. Figure 1: An overview of our decoupled evaluation protocol. We assess each atomic sub-capability independently using oracle intermediate contexts to prevent error propagation. our experiments reveal three critical observations regarding the component-wise capabilities of LLMs: • Deficits in Open-World Constraint Inference: While models demonstrate proficiency in extracting explicit, closed-world constraints and hand… view at source ↗
Figure 3
Figure 3. Figure 3: Top 10 specific confusion pairs (oracle → predicted). The results show that LLMs are prone to confusing similar tools in complex scenarios. Open-constraint and expression diversity challenges. The performance landscape shifts drastically on the Chi￾naTravel dataset, marking a precipitous drop. F1 scores fall to the 0.61–0.71 range, but most critically, the dataset yields a universal 0.00 EM rate across all… view at source ↗
Figure 4
Figure 4. Figure 4: POI match rate and coverage across different settings. The results show that LLMs tend to favor certain POIs over others when generating plans. total cost, remain within the specified budget view at source ↗
Figure 5
Figure 5. Figure 5: Error persistence on TravelPlanner. The results show that non-reasoning models and some reasoning models struggle to avoid existing errors. that the rule is violated when the number of travelers ex￾ceeds the maximum capacity of a single room, rather than recognizing that the solution could involve booking addi￾tional rooms. Similarly, for room type, the model may mistakenly assumes that a private room does… view at source ↗
Figure 6
Figure 6. Figure 6: and view at source ↗
Figure 7
Figure 7. Figure 7: Statistics of plan error types under different settings on TripCraft 12 view at source ↗
read the original abstract

Travel planning serves as a critical task for long-horizon reasoning, exposing significant deficits in LLMs. However, existing benchmarks and evaluations primarily assess final plans in an end-to-end manner, which lacks interpretability and makes it difficult to analyze the root causes of failures. To bridge this gap, we decompose travel planning into five constituent atomic sub-capabilities, including \emph{Constraint Extraction}, \emph{Tool Use}, \emph{Plan Generation}, \emph{Error Identification}, and \emph{Error Correction}. We implement a decoupled evaluation protocol leveraging oracle intermediate contexts to rigorously isolate these components, thereby measuring the atomic performance boundary without the noise of cascading errors. Our results highlight a clear contrast in performance: while LLMs are proficient in extracting explicit constraints, they struggle to infer implicit, open-world requirements. Furthermore, they exhibit structural biases in plan generation and suffer from ineffective self-correction, characterized by excessive sensitivity and erroneous persistence. These findings offer precise directions for improving LLM reasoning and planning abilities.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper decomposes travel planning into five atomic sub-capabilities (Constraint Extraction, Tool Use, Plan Generation, Error Identification, Error Correction) and introduces a decoupled evaluation protocol that supplies oracle intermediate contexts to isolate each component without cascading errors. It reports that LLMs handle explicit constraints well but struggle with implicit open-world requirements, exhibit structural biases during plan generation, and display ineffective self-correction characterized by excessive sensitivity and erroneous persistence.

Significance. If the empirical contrasts hold under more realistic conditions, the work supplies a useful fine-grained diagnostic for LLM long-horizon reasoning deficits and concrete targets for improvement. The decomposition itself and the explicit isolation of sub-capabilities constitute a methodological contribution that could be adopted by subsequent studies.

major comments (2)
  1. [Evaluation Protocol] The decoupled protocol (Abstract and Evaluation section) supplies perfect oracle outputs for prior stages when testing Error Identification and Error Correction. This removes realistic cascading mistakes, so the reported 'excessive sensitivity and erroneous persistence' in self-correction may be an artifact of the clean context rather than an intrinsic limitation; the same concern applies to the structural biases claimed for Plan Generation.
  2. [Results] No implementation details, concrete benchmarks, quantitative tables, or error analysis appear in the abstract or are referenced in the reader's summary, making it impossible to verify the magnitude of the claimed performance contrasts or to reproduce the structural-bias findings.
minor comments (1)
  1. [Abstract] The abstract would be strengthened by naming the specific LLMs, travel-planning dataset, and number of instances used so readers can immediately gauge the scope of the evaluation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful and constructive report. We address each major comment below with clarifications on our methodology and planned revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Evaluation Protocol] The decoupled protocol (Abstract and Evaluation section) supplies perfect oracle outputs for prior stages when testing Error Identification and Error Correction. This removes realistic cascading mistakes, so the reported 'excessive sensitivity and erroneous persistence' in self-correction may be an artifact of the clean context rather than an intrinsic limitation; the same concern applies to the structural biases claimed for Plan Generation.

    Authors: We appreciate this important point regarding the decoupled protocol. The design intentionally supplies oracle intermediate contexts to isolate each sub-capability and measure its atomic performance boundary without confounding from upstream errors, which is the core methodological contribution for fine-grained diagnosis. We agree that this may not fully replicate cascading effects in fully realistic end-to-end settings. In the revision, we will expand the Evaluation section with a dedicated limitations paragraph explicitly discussing this trade-off and will include additional end-to-end experiments (without oracle contexts) to show how the isolated deficits manifest under more integrated conditions. revision: partial

  2. Referee: [Results] No implementation details, concrete benchmarks, quantitative tables, or error analysis appear in the abstract or are referenced in the reader's summary, making it impossible to verify the magnitude of the claimed performance contrasts or to reproduce the structural-bias findings.

    Authors: The abstract is a concise high-level summary by design and does not contain implementation details or tables. The full manuscript provides these in Section 3 (Methodology and Implementation), Section 4 (Experiments and Benchmarks) with quantitative tables reporting performance on each sub-capability, and Section 5 (Error Analysis) that breaks down structural biases and self-correction patterns with concrete examples. The reader's summary is an external overview and not part of the paper. We will revise the abstract to include brief pointers to these sections and ensure all claims are directly supported by the presented data and released artifacts. revision: partial

Circularity Check

0 steps flagged

No circularity; empirical evaluation defines its own test protocol without reduction to fitted inputs or self-citations

full rationale

The paper decomposes travel planning into five explicitly defined sub-capabilities and implements a decoupled oracle protocol to isolate performance, as stated in the abstract: 'we decompose travel planning into five constituent atomic sub-capabilities... leveraging oracle intermediate contexts to rigorously isolate these components, thereby measuring the atomic performance boundary without the noise of cascading errors.' This methodological choice is self-contained and does not derive any result by construction from prior fits, self-citations, or renamings. All claims (e.g., proficiency in explicit constraints vs. struggles with implicit ones) are direct empirical measurements under the stated protocol, with no load-bearing self-referential steps or equations that collapse to inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Empirical benchmarking paper with no mathematical derivations, fitted parameters, or postulated entities; relies on standard assumptions about LLM evaluation.

pith-pipeline@v0.9.0 · 5488 in / 1017 out tokens · 43267 ms · 2026-05-07T16:48:23.743440+00:00 · methodology

Review history (2 revisions) →

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

38 extracted references · 14 canonical work pages · 3 internal anchors

  1. [1]

    Tripcraft: A benchmark for spatio-temporally fine grained travel planning.arXiv preprint arXiv:2502.20508,

    Chaudhuri, S., Purkar, P., Raghav, R., Mallick, S., Gupta, M., Jana, A., and Ghosh, S. Tripcraft: A benchmark for spatio-temporally fine grained travel planning.arXiv preprint arXiv:2502.20508,

    work page arXiv

  2. [2]

    Atlas: Constraints-aware multi- agent collaboration for real-world travel planning.arXiv preprint arXiv:2509.25586, 2025

    Choi, J., Yoon, J., Chen, J., Jha, S., and Pfister, T. At- las: Constraints-aware multi-agent collaboration for real- world travel planning.arXiv preprint arXiv:2509.25586,

    work page arXiv

  3. [3]

    Retail: Towards real-world travel planning for large language models

    Guo, Y ., and Wang, Y . Retail: Towards real-world travel planning for large language models. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pp. 14881–14913, 2025

    2025

  4. [3]

    Retail: Towards real-world travel planning for large language models

    Deng, B., Feng, Y ., Liu, Z., Wei, Q., Zhu, X., Chen, S., Guo, Y ., and Wang, Y . Retail: Towards real-world travel planning for large language models. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pp. 14881–14913,

    2025

  5. [4]

    Memp: Exploring Agent Procedural Memory

    Fang, R., Liang, Y ., Wang, X., Wu, J., Qiao, S., Xie, P., Huang, F., Chen, H., and Zhang, N. Memp: Exploring agent procedural memory.arXiv preprint arXiv:2508.06433,

    work page internal anchor Pith review Pith/arXiv arXiv

  6. [5]

    DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

    Guo, D., Yang, D., Zhang, H., Song, J., Zhang, R., Xu, R., Zhu, Q., Ma, S., Wang, P., Bi, X., et al. Deepseek-r1: In- centivizing reasoning capability in llms via reinforcement learning.arXiv preprint arXiv:2501.12948,

    work page internal anchor Pith review arXiv

  7. [6]

    Large language models can solve real-world planning rigorously with formal verification tools

    Hao, Y ., Chen, Y ., Zhang, Y ., and Fan, C. Large language models can solve real-world planning rigorously with formal verification tools. InProceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pp. 3434–3483, 2025

    2025

  8. [6]

    Large language models can solve real-world planning rigorously with formal verification tools

    Hao, Y ., Chen, Y ., Zhang, Y ., and Fan, C. Large language models can solve real-world planning rigorously with formal verification tools. InProceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pp. 3434–3483,

    2025

  9. [7]

    Hua, W., Wan, M., V ADREVU, J. S. S. S., Nadel, R., Zhang, Y ., and Wang, C. Interactive speculative planning: En- hance agent efficiency through co-design of system and user interface. InProceedings of the 13th International Conference on Learning Representations, 2025

    2025

  10. [7]

    T., Fazel-Zarandi, M., and Tian, Y

    Ju, D., Jiang, S., Cohen, A., Foss, A., Mitts, S., Zharmagam- betov, A., Amos, B., Li, X., Kao, J. T., Fazel-Zarandi, M., and Tian, Y . To the globe (TTG): Towards language- driven guaranteed travel planning. InProceedings of the 2024 Conference on Empirical Methods in Natural Lan- guage Processing: System Demonstrations, pp. 240–249,

    2024

  11. [8]

    P., Tafjord, O., and Clark, P

    Jansen, P., C ˆot´e, M.-A., Khot, T., Bransom, E., Dalvi Mishra, B., Majumder, B. P., Tafjord, O., and Clark, P. Discoveryworld: A virtual environment for developing and evaluating automated scientific discovery agents.Ad- vances in Neural Information Processing Systems, 2024

    2024

  12. [8]

    Triptide: A benchmark for adap- tive travel planning under disruptions.arXiv preprint arXiv:2510.21329,

    Karmakar, P., Chaudhuri, S., Mallick, S., Gupta, M., Jana, A., and Ghosh, S. Triptide: A benchmark for adap- tive travel planning under disruptions.arXiv preprint arXiv:2510.21329,

    work page arXiv

  13. [9]

    T., Fazel-Zarandi, M., and Tian, Y

    Ju, D., Jiang, S., Cohen, A., Foss, A., Mitts, S., Zharmagam- betov, A., Amos, B., Li, X., Kao, J. T., Fazel-Zarandi, M., and Tian, Y . To the globe (TTG): Towards language- driven guaranteed travel planning. InProceedings of the 2024 Conference on Empirical Methods in Natural Lan- guage Processing: System Demonstrations, pp. 240–249, 2024

    2024

  14. [9]

    Labu- topia: High-fidelity simulation and hierarchical bench- mark for scientific embodied agents.arXiv preprint arXiv:2505.22634,

    Li, R., Hu, Z., Qu, W., Zhang, J., Yin, Z., Zhang, S., Huang, X., Wang, H., Wang, T., Pang, J., Ouyang, W., Bai, L., Zuo, W., Duan, L.-Y ., Zhou, D., and Tang, S. Labu- topia: High-fidelity simulation and hierarchical bench- mark for scientific embodied agents.arXiv preprint arXiv:2505.22634,

    work page arXiv

  15. [10]

    P., and Murthy, A

    Stechly, K., Bhambri, S., Saldyt, L. P., and Murthy, A. B. Position: Llms can’t plan, but can help planning in llm- modulo frameworks. InProceedings of the 41st Inter- national Conference on Machine Learning, pp. 22895– 22907, 2024

    2024

  16. [10]

    Decompose, plan in parallel, and merge: A novel paradigm for large language models based planning with multiple constraints.arXiv preprint arXiv:2506.02683,

    Lu, Z., Lu, W., Tao, Y ., Dai, Y ., Chen, Z., Zhuang, H., Chen, C., Peng, H., and Zeng, Z. Decompose, plan in parallel, and merge: A novel paradigm for large language models based planning with multiple constraints.arXiv preprint arXiv:2506.02683,

    work page arXiv

  17. [11]

    Deeptravel: An end-to-end agen- tic reinforcement learning framework for autonomous travel planning agents.arXiv preprint arXiv:2509.21842,

    Ning, Y ., Liu, R., Wang, J., Chen, K., Li, W., Fang, J., Zheng, K., Tan, N., and Liu, H. Deeptravel: An end-to-end agen- tic reinforcement learning framework for autonomous travel planning agents.arXiv preprint arXiv:2509.21842,

    work page arXiv

  18. [12]

    2024- 09-12

    URL https://openai.com/index/ learning-to-reason-with-llms/ . 2024- 09-12. Qin, Y ., Liang, S., Ye, Y ., Zhu, K., Yan, L., Lu, Y ., Lin, Y ., Cong, X., Tang, X., Qian, B., et al. Toolllm: Facilitating large language models to master 16000+ real-world apis. InProceedings of the 12th International Conference on Learning Representations,

    2024

  19. [13]

    Trip- score: Benchmarking and rewarding real-world travel planning with fine-grained evaluation.arXiv preprint arXiv:2510.09011,

    9 Revisiting the Travel Planning Capabilities of Large Language Models Qu, Y ., Xiao, H., Li, F., Zhou, H., and Dai, X. Trip- score: Benchmarking and rewarding real-world travel planning with fine-grained evaluation.arXiv preprint arXiv:2510.09011,

    work page arXiv

  20. [14]

    Llm with tools: A survey.arXiv preprint arXiv:2409.18807,

    Shen, Z. Llm with tools: A survey.arXiv preprint arXiv:2409.18807,

    work page arXiv

  21. [15]

    Toolllm: Facilitating large language models to master 16000+ real-world apis

    Cong, X., Tang, X., Qian, B., et al. Toolllm: Facilitating large language models to master 16000+ real-world apis. InProceedings of the 12th International Conference on Learning Representations, 2024. 9 Revisiting the Travel Planning Capabilities of Large Language Models

    2024

  22. [15]

    Personal large language model agents: A case study on tailored travel planning

    Singh, H., Verma, N., Wang, Y ., Bharadwaj, M., Fashandi, H., Ferreira, K., and Lee, C. Personal large language model agents: A case study on tailored travel planning. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing: Industry Track, pp. 486–514,

    2024

  23. [16]

    github.io/IJCAI2025/

    URL https://chinatravel-competition. github.io/IJCAI2025/. 2025-08-25. Wang, K., Shen, Y ., Lv, C., Zheng, X., and Huang, X.- J. Triptailor: A real-world benchmark for personalized travel planning. InFindings of the Association for Com- putational Linguistics, pp. 9705–9723,

    2025

  24. [17]

    Chinatravel: An open-ended travel planning benchmark with compositional constraint validation for language agents

    Guo, L.-Z., and Li, Y .-F. Chinatravel: An open-ended travel planning benchmark with compositional constraint validation for language agents. InProceedings of the 14th International Conference on Learning Representations, 2026

    2026

  25. [17]

    Scienceworld: Is your agent smarter than a 5th grader? InProceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pp

    Wang, R., Jansen, P., C ˆot´e, M.-A., and Ammanabrolu, P. Scienceworld: Is your agent smarter than a 5th grader? InProceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pp. 11279– 11298,

    2022

  26. [18]

    Personal travel solver: A preference-driven llm-solver system for travel planning

    Shao, Z., Wu, J., Chen, W., and Wang, X. Personal travel solver: A preference-driven llm-solver system for travel planning. InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp. 27622–27642, 2025

    2025

  27. [18]

    2023-06-23

    URL https://lilianweng.github.io/posts/ 2023-06-23-agent/. 2023-06-23. Xie, C. and Zou, D. A human-like reasoning framework for multi-phases planning task with large language models. arXiv preprint arXiv:2405.18208,

    work page arXiv 2023

  28. [19]

    Auto-gpt for online decision making: Benchmarks and additional opinions.arXiv preprint arXiv:2306.02224,

    Yang, H., Yue, S., and He, Y . Auto-gpt for online decision making: Benchmarks and additional opinions.arXiv preprint arXiv:2306.02224,

    work page arXiv

  29. [20]

    Reflexion: Language agents with verbal rein- forcement learning.Advances in Neural Information Processing Systems, 2023

    Yao, S. Reflexion: Language agents with verbal rein- forcement learning.Advances in Neural Information Processing Systems, 2023

    2023

  30. [20]

    Agent learning via early experience.arXiv preprint arXiv:2510.08558, 2025

    Zhang, C., Goh, X. D., Li, D., Zhang, H., and Liu, Y . Plan- ning with multi-constraints via collaborative language agents. InProceedings of the 31st International Confer- ence on Computational Linguistics, pp. 10054–10082, 2025a. Zhang, K., Chen, X., Liu, B., Xue, T., Liao, Z., Liu, Z., Wang, X., Ning, Y ., Chen, Z., Fu, X., et al. Agent learning via ear...

    work page arXiv

  31. [21]

    Personal large language model agents: A case study on tailored travel planning

    Singh, H., Verma, N., Wang, Y ., Bharadwaj, M., Fashandi, H., Ferreira, K., and Lee, C. Personal large language model agents: A case study on tailored travel planning. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing: Industry Track, pp. 486–514, 2024. TPC Organizers. Chinatravel competition, 2025. URL https://china...

    2024

  32. [21]

    Statistics of Errors in Plan Generation Figure 6 and Figure 7 show the specific causes of errors in the plans generated by the model

    10 Revisiting the Travel Planning Capabilities of Large Language Models A. Statistics of Errors in Plan Generation Figure 6 and Figure 7 show the specific causes of errors in the plans generated by the model. On TripCraft, Gemini 3 Pro exhibits a high number of all error types due to the large number of solutions with formatting errors. 0 20 40 60 80 Erro...

    2022

  33. [22]

    Triptailor: A real-world benchmark for personalized travel planning

    Wang, K., Shen, Y ., Lv, C., Zheng, X., and Huang, X.- J. Triptailor: A real-world benchmark for personalized travel planning. InFindings of the Association for Com- putational Linguistics, pp. 9705–9723, 2025

    2025

  34. [23]

    Scienceworld: Is your agent smarter than a 5th grader? InProceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pp

    Wang, R., Jansen, P., C ˆot´e, M.-A., and Ammanabrolu, P. Scienceworld: Is your agent smarter than a 5th grader? InProceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pp. 11279– 11298, 2022

    2022

  35. [24]

    Llm-powered autonomous agents, 2023

    Weng, L. Llm-powered autonomous agents, 2023. URL https://lilianweng.github.io/posts/ 2023-06-23-agent/. 2023-06-23

    2023

  36. [26]

    Travelplanner: a benchmark for real-world planning with language agents

    Xie, J., Zhang, K., Chen, J., Zhu, T., Lou, R., Tian, Y ., Xiao, Y ., and Su, Y . Travelplanner: a benchmark for real-world planning with language agents. InProceedings of the 41st International Conference on Machine Learning, pp. 54590–54613, 2024

    2024

  37. [28]

    R., and Cao, Y

    Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K. R., and Cao, Y . React: Synergizing reasoning and acting in language models. InProceedings of the 11th International Conference on Learning Representations, 2023

    2023

  38. [30]

    A Survey of Large Language Models

    Min, Y ., Zhang, B., Zhang, J., Dong, Z., et al. A survey of large language models.arXiv preprint arXiv:2303.18223, 2023. 10 Revisiting the Travel Planning Capabilities of Large Language Models A. Statistics of Errors in Plan Generation Figure 6 and Figure 7 show the specific causes of errors in the plans generated by the model. On TripCraft, Gemini 3 Pro...

    work page internal anchor Pith review arXiv 2023