Ubiquant IQuest Research
Research Intern · Foundation Model Mid-training and Post-training
Contributing to STEM reasoning and agent capabilities in foundation model post-training, including IQuest-Q1.
IQuest-Q1 blog ↗ · Model & code ↗
Beijing, China
Xidong WANG
Ph.D. Candidate · CUHK-Shenzhen
I am a Ph.D. candidate in Computer Science at The Chinese University of Hong Kong, Shenzhen, advised by Prof. Benyou Wang. I received my B.Eng. from Beijing Institute of Technology as part of Xuteli School’s elite talent program.
I am currently a research intern at Ubiquant IQuest Research.
Email: xidongwang1@link.cuhk.edu.cn
Research
My goal is to democratize progress toward AGI. I develop architectural and algorithmic methods that advance model capabilities while remaining efficient enough for researchers with modest resources to explore, contribute to, and build upon:
- Hybrid architectures for scalable multimodal learning (LongLLaVA)
- Learning connections between Transformer blocks (T-Tuning, forthcoming)
- Value-guided reinforcement learning for reasoning (PPO+, forthcoming)
My earlier work pursued the same goal through medical AI for social good, building open resources to broaden access:
- Medical evaluation and leakage-resistant benchmarks (CMB, LiveClin)
- Healthcare alignment from clinician preferences (ClinAlign)
- Medical models for 50 languages (ApolloMoE)
- Open medical datasets and models (Huatuo-26M, Huatuo Series)
See publications for the full list of my research papers.
News
Scroll for more ↓- IQuest-Q1 is out! Contributed to STEM and agent post-training. Blog ↗
- Two papers accepted to NeurIPS 2026, including one Oral! Congratulations to all co-authors!
- Two papers accepted to the EMNLP 2026! Congratulations to all co-authors!
- Two papers accepted to ACL 2026! Congratulations to all co-authors!
- Joined Ubiquant IQuest Research to work on foundation model post-training for STEM reasoning and agents.
- One paper accepted to ICLR 2026: LiveClin! Congratulations to all co-authors!
- AQ became Ant Afu (蚂蚁阿福)! I contributed to the foundation model behind this AI healthcare app. Try it ↗
- ApolloMoE accepted to ICLR 2025, and LongLLaVA to EMNLP 2025!
Industry Experience
Ant Healthcare, Ant Group
Research Intern · Foundation Model Post-training
Worked on knowledge injection and rubric-based RL for Ant Afu (蚂蚁阿福)’s foundation model. The app reached 150M users (Sep. 2026) and featured in TIME’s Best Inventions of 2026 ↗.
Hangzhou, China
Shenzhen Research Institute of Big Data
Research Intern · Medical Language Models
Contributed to the early development of the HuatuoGPT series.
Shenzhen, China
Selected publications
All publications →LiveClin: A Live Clinical Benchmark without Leakage
ICLR · 2026Paper ↗Code ↗LongLLaVA: Scaling Multi-modal LLMs to 1000 Images Efficiently via a Hybrid Architecture
EMNLP · 2025Paper ↗Code ↗CMB: A Comprehensive Medical Benchmark in Chinese
NAACL · 2024Paper ↗Code ↗Website ↗Participation from publicly listed companies
ClinAlign: Scaling Healthcare Alignment from Clinician Preference
ACL · 2026Paper ↗Huatuo-26M, a Large-scale Chinese Medical QA Dataset
NAACL · 2025Paper ↗Code ↗Apollo: A Lightweight Multilingual Medical LLM towards Democratizing Medical AI to 6B People
Preprint · 2024Paper ↗Code ↗120K+ downloads on Hugging Face
Education
The Chinese University of Hong Kong, Shenzhen
Ph.D. Candidate in Computer Science
Advisor: Prof. Benyou Wang
Beijing Institute of Technology
B.Eng. in Computer Science and Technology
Xuteli School · Elite talent development
Teaching & service
CSC6052 · Natural Language Processing
Teaching Assistant · CUHK-Shenzhen
CSC6203 · Large Language Models
Teaching Assistant · CUHK-Shenzhen
Both courses taught by Prof. Benyou Wang.
Reviewing. ICLR, ICML, NeurIPS, COLM, TMLR, and ACL Rolling Review.
Open source. Lead maintainer of Awesome-AI4Med, a collection of resources for medical AI. 2.9K+ GitHub stars.