Fan Yang - Agentic AI

Agent is not a new concept, and the importance of agent in the AI era has been well recoganized. An agent can be viewed as a neural-symbolic approach, as discussed here. But I prefer to view agent as intelligence augmented by tools and their associated environments. Just as human’s capability is greatly enhanced by tools, AI’s capability grows significantly with the help of tools. Our study suggests that AI’s math reasoning can be significantly improved with the help of an MCTS-based, code augmented tool (Guan et al., 2025)(Shang et al., 2025), or a multi-agent system through a generation-discrimination process (Qi et al., 2025). More recently, we discover an agentic flow resembles the old DSP (draft, sketch, and prove) paradigm where general LLMs’ math reasoning capability can compete with LLMs trained specifically for math, even without any training (Cao et al., 2025). Based on this insight, we discover a new agentic reasoning paradigm, vibe reasoning (Wu et al., 2025), where AI reasons under the problem-agnostic guidance from human. Vibe reasoning solves IMO’25 P6, currently beyond the reach of frontier AI.

It is also worth noting that Agent will influence the future of systems. For example, will the future LLM serving systems look more like Parrot (Lin et al., 2024), or do we need to build new RL simulators like Android Arena (Xing et al., 2024) to enable OS agent?

References

2025

ICML

rStar-Math: Small LLMs Can Master Math Reasoning with Self-Evolved Deep Thinking

Xinyu Guan, and 7 more authors

ICML 2025.

Bib PAPER

ICML Oral paper (120 of 3260 accepted papers). Selected as Hugging Face daily papers: #1 paper of the day (2025-01-09).

@inproceedings{rstarmaths25,
  title = {rStar-Math: Small LLMs Can Master Math Reasoning with Self-Evolved Deep Thinking},
  author = {Guan, Xinyu and Zhang, Li Lyna and Liu, Yifei and Shang, Ning and Sun, Youran and Zhu, Yi and Yang, Fan and Yang, Mao},
  year = {2025},
  booktitle = {ICML 2025},
}

ArXiv

rStar2-Agent: Agentic Reasoning Technical Report

Ning Shang, and 14 more authors

ArXiv 2025.

Bib PAPER

@article{rstar2agent25,
  title = {rStar2-Agent: Agentic Reasoning Technical Report},
  author = {Shang, Ning and Liu, Yifei and Zhu, Yi and Zhang, Li Lyna and Xu, Weijiang and Guan, Xinyu and Zhang, Buze and Dong, Bingcheng and Zhou, Xudong and Zhang, Bowen and Xin, Ying and Miao, Ziming and Li, Scarlett and Yang, Fan and Yang, Mao},
  year = {2025},
  journal = {ArXiv},
}

ICLR

Mutual Reasoning Makes Smaller LLMs Stronger Problem-Solvers

Zhenting Qi, and 5 more authors

ICLR 2025.

Bib PAPER Code

@inproceedings{qi2025mutualreasoningmakessmaller,
  title = {Mutual Reasoning Makes Smaller LLMs Stronger Problem-Solvers},
  author = {Qi, Zhenting and Ma, Mingyuan and Xu, Jiahang and Zhang, Li Lyna and Yang, Fan and Yang, Mao},
  year = {2025},
  booktitle = {International Conference on Learning Representations, {ICLR}},
}

NeurIPS

Reviving DSP for Advanced Theorem Proving in the Era of Reasoning Models

Chenrui Cao, and 6 more authors

NeurIPS 2025.

Bib PAPER

@inproceedings{dspplus25,
  title = {Reviving DSP for Advanced Theorem Proving in the Era of Reasoning Models},
  author = {Cao, Chenrui and Song, Liangcheng and Li, Zenan and Le, Xinyi and Zhang, Xian and Xue, Hui and Yang, Fan},
  year = {2025},
  booktitle = {Advances in Neural Information Processing Systems, {NeurIPS}},
}

ArXiv

Vibe Reasoning: Eliciting Frontier AI Mathematical Capabilities – A Case Study on IMO 2025 Problem 6

Jiaao Wu, and 3 more authors

ArXiv 2025.

Bib PAPER

@article{vibereasoning25,
  title = {Vibe Reasoning: Eliciting Frontier AI Mathematical Capabilities -- A Case Study on IMO 2025 Problem 6},
  author = {Wu, Jiaao and Zhang, Xian and Yang, Fan and Dong, Yinpeng},
  year = {2025},
  journal = {ArXiv},
}

2024

OSDI

Parrot: Efficient Serving of LLM-based Applications with Semantic Variable

Chaofan Lin, and 6 more authors

OSDI 2024.

Bib PAPER Code Slides

@inproceedings{parrot24,
  title = {Parrot: Efficient Serving of LLM-based Applications with Semantic Variable},
  author = {Lin, Chaofan and Han, Zhenhua and Zhang, Chengruidong and Yang, Yuqing and Yang, Fan and Chen, Chen and Qiu, Lili},
  year = {2024},
  booktitle = {18th {USENIX} Symposium on Operating Systems Design and Implementation, {OSDI}},
}

KDD
Understanding the Weakness of Large Language Model Agents within a Complex Android Environment

Mingzhe Xing, and 5 more authors

KDD 2024, Barcelona, Spain.

Abs Bib PAPER

Large language models (LLMs) have empowered intelligent agents to execute intricate tasks within domain-specific software such as browsers and games. However, when applied to general-purpose software systems like operating systems, LLM agents face three primary challenges. Firstly, the action space is vast and dynamic, posing difficulties for LLM agents to maintain an up-to-date understanding and deliver accurate responses. Secondly, real-world tasks often require inter-application cooperation, demanding farsighted planning from LLM agents. Thirdly, agents need to identify optimal solutions aligning with user constraints, such as security concerns and preferences. These challenges motivate AndroidArena, an environment and benchmark designed to evaluate LLM agents on a modern operating system. To address high-cost of manpower, we design a scalable and semi-automated method to construct the benchmark. In the task evaluation, AndroidArena incorporates accurate and adaptive metrics to address the issue of non-unique solutions. Our findings reveal that even state-of-the-art LLM agents struggle in cross-APP scenarios and adhering to specific constraints. Additionally, we identify a lack of four key capabilities, i.e. understanding, reasoning, exploration, and reflection, as primary reasons for the failure of LLM agents. Furthermore, we provide empirical analysis on the failure of reflection, and improve the success rate by 27% with our proposed exploration strategy. This work is the first to present valuable insights in understanding fine-grained weakness of LLM agents, and offers a path forward for future research in this area. Environment, benchmark, prompt, and evaluation code for AndroidArena are released at https://github.com/AndroidArenaAgent/AndroidArena.
@inproceedings{10.1145/3637528.3671650, title = {Understanding the Weakness of Large Language Model Agents within a Complex Android Environment}, author = {Xing, Mingzhe and Zhang, Rongkai and Xue, Hui and Chen, Qi and Yang, Fan and Xiao, Zhen}, year = {2024}, booktitle = {Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining}, location = {Barcelona, Spain}, publisher = {Association for Computing Machinery}, address = {New York, NY, USA}, series = {KDD '24}, pages = {6061–6072}, doi = {10.1145/3637528.3671650}, isbn = {9798400704901}, numpages = {12}, keywords = {ai agent, large language model, task planning} }