Yuyao Ge 葛钰峣

Yuyao Ge 葛钰峣

Researcher · Backpacker

Photographer · Hiker

Institute of Computing Technology (ICT)

Chinese Academy of Sciences (CAS)

About

FIELD: AGENTIC INTELLIGENCE · BEIJING · 2026

Hi👋, welcome to my website! I am a Ph.D. Student at Institute of Computing Technology, Chinese Academy of Sciences, advised by Prof. Shenghua Liu and Prof. Yiwei Wang. I work on Long-Horizon LLM Reasoning and Agentic Code Generation.

I am currently a Research Intern at Moonshot AI, where I participated in the Foundation Model Training for Kimi K2, K2.5, and K2.6, and explored the Coding capabilities of the K2 series models. Previously, I interned at Microsoft Research Asia and Tsinghua University.

Outside of work, I’m a backpacker with interests in hiking ⛰️ and photography 🎞️. You can view my photography work HERE.

Interests
  • Large Language Model Reasoning
  • Long-Horizon Agent
  • Agentic Code Generation
  • Context Management
Education

  • Ph.D Candidate, 2024-Present

    Institute of Computing Technology

  • B.S. in Computer Science, 2020-2024

    North China University of Technology

Highlight Papers

Kimi K2.6: Advancing Open-Source Coding
We are open sourcing our latest model, Kimi K2.6, featuring state-of-the-art coding, long-horizon execution, and agent swarm …
Cite Blog Hugging Face Ask KIMI
Kimi K2.6: Advancing Open-Source Coding
Kimi K2.5: Visual Agentic Intelligence
We introduce Kimi K2.5, an open-source multimodal agentic model designed to advance general agentic intelligence. The model focuses on …
Cite PDF arXiv Blog Hugging Face GitHub Ask KIMI
Kimi K2.5: Visual Agentic Intelligence
A Survey of Vibe Coding with Large Language Models
The advancement of large language models (LLMs) has catalyzed a paradigm shift from code generation assistance to autonomous coding …
Cite Wiki PDF arXiv Hugging Face Ask KIMI
A Survey of Vibe Coding with Large Language Models
Focusing by Contrastive Attention: Enhancing VLMs' Visual Reasoning
Vision-Language Models (VLMs) have demonstrated remarkable success across diverse visual tasks, yet their performance degrades in …
Cite PDF arXiv Ask KIMI PaperWeekly
Focusing by Contrastive Attention: Enhancing VLMs' Visual Reasoning
Can Graph Descriptive Order Affect Solving Graph Problems with LLMs?
We present the first comprehensive analysis of how the order of graph descriptions impacts LLM performance, evaluating four graph description orders across six graph problems using six mainstream LLMs.
Cite Slides Video PDF ACL Anthology Ask KIMI Poster GitHub
Can Graph Descriptive Order Affect Solving Graph Problems with LLMs?
Kimi K2: Open Agentic Intelligence
We introduce Kimi K2, a Mixture-of-Experts large language model with 32 billion activated parameters and 1 trillion total parameters. …
Cite PDF arXiv Blog Hugging Face GitHub Ask KIMI
Kimi K2: Open Agentic Intelligence
A Survey of Context Engineering for Large Language Models
The performance of Large Language Models (LLMs) is fundamentally determined by the contextual information provided during inference. …
Lingrui Mei, Jiayu Yao, Yuyao Ge 葛钰峣, Yiwei Wang, Baolong Bi, Yujun Cai, Jiazhi Liu, Mingyu Li, Zhong-Zhi Li, Duzhen Zhang, Chenlin Zhou, Jiayi Mao, Tianze Xia, Jiafeng Guo, Shenghua Liu
Cite PDF arXiv Hugging Face GitHub Ask KIMI
A Survey of Context Engineering for Large Language Models
Innate Reasoning is Not Enough : In-Context Learning Enhances Reasoning Large Language Models with Less Overthinking
We present the first comprehensive analysis of the impacts of CoT prompting on Reasoning LLMs, finding that one-shot CoT consistently enhances performance and reduces excessive reflections by approximately 90%.
Cite PDF arXiv Hugging Face Ask KIMI
Innate Reasoning is Not Enough : In-Context Learning Enhances Reasoning Large Language Models with Less Overthinking
Attack based on data : A novel perspective to attack sensitive points directly
Adversarial attack for time-series classification model is widely explored and many attack methods are proposed. But there is not a …
Yuyao Ge 葛钰峣, Zhongguo Yang, Lizhe Chen, Yiming Wang, Chengyang Li
Cite Dataset PDF Page Ask KIMI
Attack based on data : A novel perspective to attack sensitive points directly

Experience

 
 
 
 
 
Hunyuan Team, Tencent
Research Intern
Hunyuan Team, Tencent
May 2026 – Present Beijing, China
 
 
 
 
 
RL Team, Moonshot AI
Research Intern
RL Team, Moonshot AI
September 2025 – May 2026 Beijing, China
Coding Ability of Kimi K2 Improved Kimi K2's coding capabilities through reinforcement learning Context Management of Kimi K2.5 Contributed context management capabilities to Kimi K2.5
Long-horizon execution of KIMI K2.6 Research on scaling long-horizon execution of KIMI K2.6
Research on Coding Ability of LLMs Research on efficient and truthful vibe coding workflows with LLMs
 
 
 
 
 
Microsoft Research Asia
Research Intern
Microsoft Research Asia
March 2024 – September 2024 Beijing, China
Deep Research of Bing Generative Search Contributed to building deep research capabilities for Bing's generative search system
Next-generation Recommendation System of Bing Trained next-generation recommendation models for Bing
 
 
 
 
 
Tsinghua University
Machine Learning Engineer Intern
Tsinghua University
September 2023 – March 2024 Beijing, China
Research on LLM Reasoning Investigated how graph descriptive order affects LLM reasoning performance LLM Post-Training Trained LLMs with agentic capabilities including search and tool use
 
 
 
 
 
PaddlePaddle, Baidu
Software Engineer Intern
PaddlePaddle, Baidu
July 2021 – December 2021 Beijing, China
Contributions to PaddlePaddle Contributed open-source code to the PaddlePaddle deep learning framework
Champion of PaddlePaddle Hackathon Built a deep-learning project at top speed in 48h — without vibe coding

Projects

Long-Insight: Long-running Agent Trajectory Analysis
Interactive analysis platform for ultra-long agent trajectories (400+ turns, 1M+ tokens). Decomposes massive coding agent trajectories into hierarchical step DAGs, scores trajectory quality, and renders interactive visualizations.
Live Demo GitHub
Long-Insight: Long-running Agent Trajectory Analysis
MIX-MCP (Eth-Beijing Runner-Up🥈)
MIX-MCP is a Go-based aggregator of MCP services aiming to build a foundational layer for interaction between large language models and Web3 technologies.
Video Live Demo GitHub Doc Eth-Beijing Pharos
MIX-MCP (Eth-Beijing Runner-Up🥈)
PHOTOGRAPHY · FIELD NOTES · 2026

Mountains · Sea · Forest

Wilderness

Landscapes that exist beyond human scale — nature indifferent to our presence.

Start Exploring

Ancient Ruins · Architecture

Traces

What humans pressed into stone and concrete — crystallized intention outlasting its makers.

Start Exploring

Wildlife · Candid Moments

Moment

Living systems running their own computations — creatures that look back.

Start Exploring

Recent Posts

更复杂的 Agent 能带来更好的性能吗？
mini-swe-agent SWE-agent SWE-bench Pro
利用 SKILL让 Agent 自动筛选 & 解读 Huggingface 的每日论文
SKILL WebSite Ask KIMI
EMNLP2024论文分享 | Fewer is More：CoT示例要少而精
编码实践 | 一文读懂Tokenizer | 以Word-based 和 Byte Pair Encoding为例
论文分享 | 广泛的解码策略导致大模型越狱
See all posts