Hao Li (李昊)
I‘m one of the select few members of technical staff @ ERNIE Team, Baidu driving the Reinforcement Learning and On-policy Distillation for our flagship model releases.
Prior to this, I was a Post-doc @ Imperial College London and Research Intern @ Microsoft Research
My research interests lie in Post-training and Reinforcement Learning. Recently, focusing on:
- Agentic RL & Reward Modeling
- Training-Inference Mismatch
- On-policy Distillation
Email / LinkedIn / Google Scholar / GitHub
Blog
ERNIE 5.1: The Next Generation Foundation Model
Announcing the release of ERNIE 5.1. Exploring the latest advancements in post-training, reinforcement learning, and our continued scaling efforts.
ERNIE 5.0: Pushing the Boundaries of Foundation Models
A deep dive into the architecture and training methodologies behind ERNIE 5.0, achieving top-tier performance on global benchmarks through advanced reinforcement learning techniques.
MIRA: Medical Time Series Foundation Model
Presenting our work at NeurIPS 2025. This paper explores MIRA, a foundation model specifically designed to tackle the complexities of real-world medical time-series data and electronic health records.
TimeCraft: A Universal Framework for Time-Series Generation
Introducing TimeCraft, a controllable generative engine for universal time-series data. Exploring the core architecture and its impact on modeling real-world sequential data.
Selected Work
ERNIE Team, Baidu
🏆 #1 in China (LMArena)
🌎 #8 Globally
🤗 2.4T(2400B) LLM
ERNIE Team, Baidu
7.6k+ Stars
👥 430M+ Users🚀 1.5B Daily Calls
Microsoft Research
1.1k+ Stars
Selected Publications
Arg-LLaDA: Argument Summarization via Large Language Diffusion Models and Sufficiency-Aware Refinement
ACL 2026
[Paper]
MIRA: Medical Time Series Foundation Model for Real-World Health Data
NeurIPS 2025
BRIDGE: Bootstrapping Text to Control Time-series Generation via Multi-agent Iterative Optimization and Diffusion Modeling
ICML 2025
TarDiff: Target-Oriented Diffusion Guidance for Synthetic Electronic Health Record Time Series Generation
KDD 2025
Does Acceleration Cause Hidden Instability in Vision Language Models? Uncovering Instance-Level Divergence Through a Large-Scale Empirical Study
EMNLP 2025
[Paper]
LVPruning: An Effective yet Simple Language-Guided Vision Token Pruning Approach for Multi-modal Large Language Models
NAACL 2025 Findings
[Paper]
Which Side Are You On? A Multi-task Dataset for End-to-End Argument Summarisation and Evaluation
ACL 2024 Findings
CIF-Bench: A Chinese Instruction-Following Benchmark for Evaluating the Generalizability of Large Language Models
ACL 2024 Findings
Do You Hear the People Sing? Key Point Analysis via Iterative Clustering and Abstractive Summarisation
ACL 2023
Not All Quantifiers Are Equal: Probing Transformer-based Language Models' Understanding of Generalised Quantifiers
EMNLP 2023
[Paper]