pith. machine review for the scientific record. sign in

arxiv: 2310.01415 · v3 · submitted 2023-10-02 · 💻 cs.CV · cs.AI· cs.CL· cs.RO

Recognition: 2 theorem links

· Lean Theorem

GPT-Driver: Learning to Drive with GPT

Jiageng Mao , Yuxi Qian , Junjie Ye , Hang Zhao , Yue Wang
Authors on Pith no claims yet

Pith reviewed 2026-05-15 14:59 UTC · model grok-4.3

classification 💻 cs.CV cs.AIcs.CLcs.RO
keywords motion planninglarge language modelsautonomous drivingtrajectory generationGPT-3.5nuScenes dataset
0
0 comments X

The pith

Reformulating motion planning as language modeling lets GPT-3.5 generate precise driving trajectories from scene descriptions.

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

The paper shows that a standard LLM can be turned into an autonomous-vehicle motion planner by recasting the entire task as next-token prediction over coordinate strings. Scene data, map information, and vehicle state are serialized into language prompts; the model then outputs a sequence of future (x, y) points together with a natural-language rationale. A three-stage prompting-reasoning-finetuning procedure is used to elicit numerical accuracy and safety reasoning. On the nuScenes benchmark the resulting trajectories are competitive with specialized planners while also supplying human-readable explanations of each decision.

Core claim

Motion planning reduces to language modeling when both inputs and outputs are expressed as token sequences of coordinate positions; an appropriately prompted and fine-tuned GPT-3.5 model therefore produces collision-free, comfortable trajectories together with explicit reasoning traces.

What carries the argument

The prompting-reasoning-finetuning pipeline that converts driving scenes into language tokens and elicits precise trajectory coordinates plus decision explanations from the LLM.

If this is right

  • The planner can output both a trajectory and a step-by-step textual justification for each choice.
  • Generalization to novel scenarios improves because the model draws on broad language-based priors rather than hand-crafted heuristics.
  • Integration into existing stacks is simplified: the same model can accept text, map, or sensor-derived prompts without custom feature engineering.

Where Pith is reading between the lines

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

  • Pairing the language planner with a separate low-level controller or safety filter could mitigate residual hallucination risks.
  • The same tokenization strategy might transfer to other continuous-control domains such as robotic arm planning or drone navigation.
  • Multimodal extensions that feed raw camera or LiDAR tokens directly into the LLM could remove the need for explicit scene serialization.

Load-bearing premise

An LLM fine-tuned only on language descriptions of coordinates will output numerically accurate, collision-free trajectories in safety-critical driving scenes it has never seen.

What would settle it

Run the model on a held-out set of nuScenes scenes containing rare maneuvers or adverse weather and measure the fraction of trajectories that intersect obstacles or violate comfort constraints.

read the original abstract

We present a simple yet effective approach that can transform the OpenAI GPT-3.5 model into a reliable motion planner for autonomous vehicles. Motion planning is a core challenge in autonomous driving, aiming to plan a driving trajectory that is safe and comfortable. Existing motion planners predominantly leverage heuristic methods to forecast driving trajectories, yet these approaches demonstrate insufficient generalization capabilities in the face of novel and unseen driving scenarios. In this paper, we propose a novel approach to motion planning that capitalizes on the strong reasoning capabilities and generalization potential inherent to Large Language Models (LLMs). The fundamental insight of our approach is the reformulation of motion planning as a language modeling problem, a perspective not previously explored. Specifically, we represent the planner inputs and outputs as language tokens, and leverage the LLM to generate driving trajectories through a language description of coordinate positions. Furthermore, we propose a novel prompting-reasoning-finetuning strategy to stimulate the numerical reasoning potential of the LLM. With this strategy, the LLM can describe highly precise trajectory coordinates and also its internal decision-making process in natural language. We evaluate our approach on the large-scale nuScenes dataset, and extensive experiments substantiate the effectiveness, generalization ability, and interpretability of our GPT-based motion planner. Code is now available at https://github.com/PointsCoder/GPT-Driver.

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 / 2 minor

Summary. The manuscript presents GPT-Driver, which reformulates autonomous-vehicle motion planning as a language-modeling task: scene inputs and trajectory outputs are tokenized as language descriptions of coordinates, and GPT-3.5 is adapted via a prompting-reasoning-finetuning pipeline to generate both trajectories and natural-language explanations of its decisions. The approach is evaluated on the nuScenes dataset, with claims of improved effectiveness, generalization to novel scenarios, and interpretability relative to heuristic planners.

Significance. If the numerical accuracy and safety claims hold under rigorous verification, the work would demonstrate a viable path for deploying LLMs in metric-sensitive planning tasks, potentially improving generalization beyond traditional methods. The public code release supports reproducibility and further investigation of the prompting-reasoning-finetuning strategy.

major comments (2)
  1. [§4 (Experiments)] §4 (Experiments) and abstract: the reported nuScenes results claim effectiveness and generalization without error bars, failure-case analysis, or explicit confirmation that generated trajectories are not post-processed. These omissions are load-bearing for the safety-critical claims in §1, as coordinate hallucinations or rounding artifacts could violate lane boundaries or obstacle clearances in OOD scenes.
  2. [§3 (Method)] §3 (Method): the prompting-reasoning-finetuning pipeline contains no post-generation projection, constraint solver, or numerical error bound to enforce metric precision or hard safety invariants. Token-level next-token prediction alone does not guarantee collision-free outputs; the manuscript therefore leaves the central assumption—that language modeling yields reliable continuous trajectories—unsupported by any compensating mechanism.
minor comments (2)
  1. [Abstract] The abstract states that the LLM can 'describe highly precise trajectory coordinates' without defining the precision metric (e.g., L2 error thresholds or collision-rate bounds) used to support this claim.
  2. [§3 (Method)] Notation for coordinate tokenization and the exact format of the language description of trajectories should be formalized with an equation or pseudocode example in §3 to improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the constructive feedback on our manuscript. We address each major comment below and indicate the revisions we will make to strengthen the paper.

read point-by-point responses

  1. Referee: [§4 (Experiments)] §4 (Experiments) and abstract: the reported nuScenes results claim effectiveness and generalization without error bars, failure-case analysis, or explicit confirmation that generated trajectories are not post-processed. These omissions are load-bearing for the safety-critical claims in §1, as coordinate hallucinations or rounding artifacts could violate lane boundaries or obstacle clearances in OOD scenes.

    Authors: We agree that error bars, failure-case analysis, and explicit confirmation of no post-processing are important for supporting the safety claims. In the revised version, we will add standard deviations from multiple runs to the key metrics in §4 and the abstract, include a new subsection analyzing representative failure cases (including OOD scenarios), and explicitly state that generated trajectories are used directly without any post-processing or projection steps. These changes will be incorporated to address the concerns. revision: yes

  2. Referee: [§3 (Method)] §3 (Method): the prompting-reasoning-finetuning pipeline contains no post-generation projection, constraint solver, or numerical error bound to enforce metric precision or hard safety invariants. Token-level next-token prediction alone does not guarantee collision-free outputs; the manuscript therefore leaves the central assumption—that language modeling yields reliable continuous trajectories—unsupported by any compensating mechanism.

    Authors: We acknowledge that the pipeline relies on the LLM's learned behavior from the prompting-reasoning-finetuning process without explicit post-generation constraints or error bounds. Our empirical results on nuScenes demonstrate that this yields safe and precise trajectories in practice, but we agree there is no theoretical guarantee against hallucinations or violations. In the revision, we will expand the discussion in §3 to explicitly note this limitation and suggest future integration of safety mechanisms, while preserving the core language-modeling approach. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical LLM fine-tuning for trajectory generation

full rationale

The paper's core contribution is an empirical reformulation of motion planning as next-token prediction on coordinate language tokens, followed by prompting-reasoning-finetuning of GPT-3.5 and evaluation on nuScenes. No mathematical derivation chain exists that reduces outputs to inputs by construction. No self-definitional equations, fitted parameters renamed as predictions, or load-bearing self-citations that justify uniqueness. The prompting strategy and coordinate tokenization are methodological choices whose validity is tested externally via dataset performance rather than assumed tautologically. This is a standard supervised fine-tuning pipeline whose results are falsifiable against held-out driving scenes.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The approach assumes that GPT-3.5 possesses latent numerical-reasoning capacity that can be elicited by language prompts and modest fine-tuning; no new physical entities or forces are postulated.

free parameters (1)
  • fine-tuning hyperparameters and prompt templates
    The strategy relies on choices of learning rate, number of epochs, and exact prompt wording that are fitted or selected to make the model output usable coordinates.
axioms (1)
  • domain assumption Large language models can be prompted to perform step-by-step numerical reasoning about spatial coordinates when given scene descriptions.
    Invoked in the description of the prompting-reasoning-finetuning strategy.

pith-pipeline@v0.9.0 · 5544 in / 1302 out tokens · 57861 ms · 2026-05-15T14:59:41.179135+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 22 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. C-TRAIL: A Commonsense World Framework for Trajectory Planning in Autonomous Driving

    cs.AI 2026-03 unverdicted novelty 7.0

    C-TRAIL combines LLM commonsense with a dual-trust mechanism and Dirichlet-weighted Monte Carlo Tree Search to improve trajectory planning accuracy and safety in autonomous driving.

  2. ReCogDrive: A Reinforced Cognitive Framework for End-to-End Autonomous Driving

    cs.CV 2025-06 unverdicted novelty 7.0

    ReCogDrive unifies VLM scene understanding with a diffusion planner reinforced by DiffGRPO to reach state-of-the-art results on NAVSIM and Bench2Drive benchmarks.

  3. MAPLE: Latent Multi-Agent Play for End-to-End Autonomous Driving

    cs.RO 2026-05 unverdicted novelty 6.0

    MAPLE performs closed-loop multi-agent training of VLA driving models entirely in latent space using supervised fine-tuning followed by RL with safety, progress, and diversity rewards, reaching SOTA on Bench2Drive.

  4. Driving Intents Amplify Planning-Oriented Reinforcement Learning

    cs.RO 2026-05 unverdicted novelty 6.0

    DIAL uses intent-conditioned CFG and multi-intent GRPO to expand and preserve diverse modes in continuous-action preference RL, lifting RFS to 9.14 and surpassing both prior best (8.5) and human demonstration (8.13).

  5. CoWorld-VLA: Thinking in a Multi-Expert World Model for Autonomous Driving

    cs.CV 2026-05 unverdicted novelty 6.0

    CoWorld-VLA encodes world information into four expert tokens that condition a diffusion-based planner, yielding competitive collision avoidance and trajectory accuracy on the NAVSIM benchmark.

  6. CoWorld-VLA: Thinking in a Multi-Expert World Model for Autonomous Driving

    cs.CV 2026-05 unverdicted novelty 6.0

    CoWorld-VLA extracts semantic, geometric, dynamic, and trajectory expert tokens from multi-source supervision and feeds them into a diffusion-based hierarchical planner, achieving competitive collision avoidance and t...

  7. ST-Prune: Training-Free Spatio-Temporal Token Pruning for Vision-Language Models in Autonomous Driving

    cs.CV 2026-04 unverdicted novelty 6.0

    ST-Prune is a training-free spatio-temporal token pruning framework for VLMs in autonomous driving that achieves near-lossless results at 90% token reduction by exploiting motion volatility, temporal recency, and mult...

  8. FeaXDrive: Feasibility-aware Trajectory-Centric Diffusion Planning for End-to-End Autonomous Driving

    cs.RO 2026-04 unverdicted novelty 6.0

    FeaXDrive improves end-to-end autonomous driving by shifting diffusion planning to a trajectory-centric formulation with curvature-constrained training, drivable-area guidance, and GRPO post-training, yielding stronge...

  9. LMGenDrive: Bridging Multimodal Understanding and Generative World Modeling for End-to-End Driving

    cs.CV 2026-04 unverdicted novelty 6.0

    LMGenDrive unifies LLM-based multimodal understanding with generative world models to output both future driving videos and control signals for end-to-end closed-loop autonomous driving.

  10. ICR-Drive: Instruction Counterfactual Robustness for End-to-End Language-Driven Autonomous Driving

    cs.CL 2026-04 unverdicted novelty 6.0

    ICR-Drive reveals substantial performance drops in end-to-end language-driven driving models when instructions are paraphrased, made ambiguous, noised, or misleading.

  11. DVGT-2: Vision-Geometry-Action Model for Autonomous Driving at Scale

    cs.CV 2026-04 unverdicted novelty 6.0

    DVGT-2 is a streaming vision-geometry-action model that jointly reconstructs dense 3D geometry and plans trajectories online, achieving better reconstruction than prior batch methods while transferring directly to pla...

  12. DRIV-EX: Counterfactual Explanations for Driving LLMs

    cs.CL 2026-02 unverdicted novelty 6.0

    DRIV-EX generates fluent counterfactual scene descriptions by using gradient-optimized embeddings only as a guide for controlled text decoding, producing more reliable explanations than baselines on transcribed highD ...

  13. AutoVLA: A Vision-Language-Action Model for End-to-End Autonomous Driving with Adaptive Reasoning and Reinforcement Fine-Tuning

    cs.CV 2025-06 unverdicted novelty 6.0

    AutoVLA unifies semantic reasoning and trajectory planning in one autoregressive VLA model for end-to-end autonomous driving by tokenizing trajectories into discrete actions and using GRPO reinforcement fine-tuning to...

  14. Driving Intents Amplify Planning-Oriented Reinforcement Learning

    cs.RO 2026-05 unverdicted novelty 5.0

    DIAL expands continuous-action driving policies via intent-conditioned flow matching and multi-intent GRPO, lifting best-of-N preference scores above human demonstrations for the first time on WOD-E2E.

  15. Latency Analysis and Optimization of Alpamayo 1 via Efficient Trajectory Generation

    cs.AI 2026-05 unverdicted novelty 5.0

    Redesigning Alpamayo 1 to single-reasoning and optimizing diffusion action generation cuts inference latency by 69.23% while preserving trajectory diversity and prediction quality.

  16. SwarmDrive: Semantic V2V Coordination for Latency-Constrained Cooperative Autonomous Driving

    cs.RO 2026-04 unverdicted novelty 5.0

    SwarmDrive uses local SLMs on vehicles for event-triggered semantic V2V intent sharing and consensus, improving occluded intersection success from 68.9% to 94.1% and cutting latency to 151.4 ms in a 5-seed simulation.

  17. SpanVLA: Efficient Action Bridging and Learning from Negative-Recovery Samples for Vision-Language-Action Model

    cs.CV 2026-04 unverdicted novelty 5.0

    SpanVLA reduces action generation latency via flow-matching conditioned on history and improves robustness by training on negative-recovery samples with GRPO and a dedicated reasoning dataset.

  18. SpaceMind: A Modular and Self-Evolving Embodied Vision-Language Agent Framework for Autonomous On-orbit Servicing

    cs.RO 2026-04 unverdicted novelty 5.0

    SpaceMind is a self-evolving modular VLM agent framework that achieves 90-100% navigation success in nominal conditions and recovers from failures via experience distillation, with zero-code transfer to physical robot...

  19. Learning Probabilistic Responsibility Allocations for Multi-Agent Interactions

    cs.MA 2026-04 unverdicted novelty 5.0

    A CVAE-based approach learns distributions over responsibility allocations in multi-agent scenes by grounding them in induced controls through differentiable optimization, showing strong prediction on driving data.

  20. EvoDriveVLA: Evolving Driving VLA Models via Collaborative Perception-Planning Distillation

    cs.CV 2026-03 unverdicted novelty 5.0

    EvoDriveVLA uses collaborative perception-planning distillation with self-anchor and future-aware teachers to fix perception degradation and long-term instability in driving VLA models, reaching SOTA on nuScenes and NAVSIM.

  21. DeepSight: Long-Horizon World Modeling via Latent States Prediction for End-to-End Autonomous Driving

    cs.CV 2026-05 unverdicted novelty 4.0

    DeepSight uses parallel latent feature prediction in BEV for long-horizon world modeling and adaptive text reasoning to reach state-of-the-art closed-loop performance on the Bench2drive benchmark.

  22. XEmbodied: A Foundation Model with Enhanced Geometric and Physical Cues for Large-Scale Embodied Environments

    cs.CV 2026-04 unverdicted novelty 4.0

    XEmbodied is a foundation model that integrates 3D geometric and physical signals into VLMs using a 3D Adapter and Efficient Image-Embodied Adapter, plus progressive curriculum and RL post-training, to improve spatial...

Reference graph

Works this paper leans on

17 extracted references · 17 canonical work pages · cited by 20 Pith papers · 7 internal anchors

  1. [1]

    Do As I Can, Not As I Say: Grounding Language in Robotic Affordances

    Michael Ahn, Anthony Brohan, Noah Brown, Yevgen Chebotar, Omar Cortes, Byron David, Chelsea Finn, Chuyuan Fu, Keerthana Gopalakrishnan, Karol Hausman, et al. Do as i can, not as i say: Grounding language in robotic affordances. arXiv preprint arXiv:2204.01691,

    work page internal anchor Pith review Pith/arXiv arXiv

  2. [2]

    End to End Learning for Self-Driving Cars

    Mariusz Bojarski, Davide Del Testa, Daniel Dworakowski, Bernhard Firner, Beat Flepp, Prasoon Goyal, Lawrence D Jackel, Mathew Monfort, Urs Muller, Jiakai Zhang, et al. End to end learning for self-driving cars. arXiv preprint arXiv:1604.07316,

    work page internal anchor Pith review Pith/arXiv arXiv

  3. [3]

    RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control

    Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen Chebotar, Xi Chen, Krzysztof Choroman- ski, Tianli Ding, Danny Driess, Avinava Dubey, Chelsea Finn, et al. Rt-2: Vision-language-action models transfer web knowledge to robotic control. arXiv preprint arXiv:2307.15818,

    work page internal anchor Pith review Pith/arXiv arXiv

  4. [4]

    Language models are few-shot learners

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901,

    work page 1901

  5. [5]

    Pix2seq: A language modeling framework for object detection

    Ting Chen, Saurabh Saxena, Lala Li, David J Fleet, and Geoffrey Hinton. Pix2seq: A language modeling framework for object detection. arXiv preprint arXiv:2109.10852,

    work page arXiv

  6. [6]

    End- to-end driving via conditional imitation learning

    Felipe Codevilla, Matthias M ¨uller, Antonio L´opez, Vladlen Koltun, and Alexey Dosovitskiy. End- to-end driving via conditional imitation learning. In 2018 IEEE international conference on robotics and automation (ICRA), pp. 4693–4700. IEEE,

    work page 2018

  7. [7]

    Parting with misconcep- tions about learning-based vehicle motion planning

    Daniel Dauner, Marcel Hallgarten, Andreas Geiger, and Kashyap Chitta. Parting with misconcep- tions about learning-based vehicle motion planning. arXiv preprint arXiv:2306.07962,

    work page arXiv

  8. [8]

    Baidu Apollo EM Motion Planner

    Haoyang Fan, Fan Zhu, Changchun Liu, Liangliang Zhang, Li Zhuang, Dong Li, Weicheng Zhu, Jiangtao Hu, Hongye Li, and Qi Kong. Baidu apollo em motion planner. arXiv preprint arXiv:1807.08048,

    work page internal anchor Pith review Pith/arXiv arXiv

  9. [9]

    Drive like a human: Rethinking autonomous driving with large language models

    Daocheng Fu, Xin Li, Licheng Wen, Min Dou, Pinlong Cai, Botian Shi, and Yu Qiao. Drive like a human: Rethinking autonomous driving with large language models. arXiv preprint arXiv:2307.07162,

    work page arXiv

  10. [10]

    Planning-oriented autonomous driving

    10 Foundation Models for Decision Making Workshop at NeurIPS 2023 Yihan Hu, Jiazhi Yang, Li Chen, Keyu Li, Chonghao Sima, Xizhou Zhu, Siqi Chai, Senyao Du, Tianwei Lin, Wenhai Wang, et al. Planning-oriented autonomous driving. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 17853–17862,

    work page 2023

  11. [11]

    Inner Monologue: Embodied Reasoning through Planning with Language Models

    Wenlong Huang, Fei Xia, Ted Xiao, Harris Chan, Jacky Liang, Pete Florence, Andy Zeng, Jonathan Tompson, Igor Mordatch, Yevgen Chebotar, et al. Inner monologue: Embodied reasoning through planning with language models. arXiv preprint arXiv:2207.05608,

    work page internal anchor Pith review Pith/arXiv arXiv

  12. [12]

    Vad: Vectorized scene representation for efficient au- tonomous driving

    Bo Jiang, Shaoyu Chen, Qing Xu, Bencheng Liao, Jiajie Chen, Helong Zhou, Qian Zhang, Wenyu Liu, Chang Huang, and Xinggang Wang. Vad: Vectorized scene representation for efficient au- tonomous driving. arXiv preprint arXiv:2303.12077,

    work page arXiv

  13. [13]

    Per- ceive, predict, and plan: Safe motion planning through interpretable semantic representations

    Abbas Sadat, Sergio Casas, Mengye Ren, Xinyu Wu, Pranaab Dhawan, and Raquel Urtasun. Per- ceive, predict, and plan: Safe motion planning through interpretable semantic representations. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXIII 16, pp. 414–430. Springer,

    work page 2020

  14. [14]

    Sadler, Wei-Lun Chao, and Yu Su

    Chan Hee Song, Jiaman Wu, Clayton Washington, Brian M Sadler, Wei-Lun Chao, and Yu Su. Llm- planner: Few-shot grounded planning for embodied agents with large language models. arXiv preprint arXiv:2212.04088,

    work page arXiv

  15. [15]

    LLaMA: Open and Efficient Foundation Language Models

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timoth ´ee Lacroix, Baptiste Rozi `ere, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023a. 11 Foundation Models for Decision Making Workshop at NeurIPS 2023 Hugo Touvron, Louis Martin...

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  16. [16]

    Visionllm: Large language model is also an open-ended decoder for vision-centric tasks

    Wenhai Wang, Zhe Chen, Xiaokang Chen, Jiannan Wu, Xizhou Zhu, Gang Zeng, Ping Luo, Tong Lu, Jie Zhou, Yu Qiao, et al. Visionllm: Large language model is also an open-ended decoder for vision-centric tasks. arXiv preprint arXiv:2305.11175,

    work page arXiv

  17. [17]

    ReAct: Synergizing Reasoning and Acting in Language Models

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. React: Synergizing reasoning and acting in language models. arXiv preprint arXiv:2210.03629,

    work page internal anchor Pith review Pith/arXiv arXiv