Systems researcher who started Chatbot Arena as a Berkeley PhD 'weekend project' and now runs Arena's technology. He built the infrastructure (FastChat, Vicuna, MT-Bench, Arena-Hard) that made crowdsourced pairwise preference the default public yardstick for models.
high confidenceSan Francisco Bay AreaAt Arena (formerly LMArena / Chatbot Arena; Arena Intelligence Inc.) since 2025Updated 2026-09-19
Background
PhD student in Ion Stoica's Sky Computing Lab at UC Berkeley, originally working on systems (SkyPilot, RL for database optimisation). After ChatGPT launched he led Vicuna, an open chatbot trained on shared ChatGPT conversations, and then Chatbot Arena to compare it against other models; he describes the team then as 'a bunch of PhD students' building a weekend prototype.
Before Berkeley he interned at Microsoft, Microsoft Research Asia, Alibaba, Google Research (Cluster-GCN) and Amazon. Felicis credits him with the decision to turn the academic project into a company.
What they run now
Owns the platform and data pipeline behind the leaderboard and paid evaluations: scaling to tens of millions of users, new modalities (video, agents), automated preference prediction and the routing product.
The latest DeepSeek release is underhyped.
DeepSeek-V4-Flash pushes the intelligence/$ frontier forward by an order of magnitude, and it's open-source. This will bring powerful agentic capabilities to the mass market and unlock 10x more use cases overnight.
Three key dimensions to scale frontier AI evals
- quality
- speed
- cost
Autoevals give you best of all worlds.
Fast, efficient, and high-quality, powered by millions of real-world tasks in Arena.
Accounts whose posts Wei-Lin has reposted recently: Arena, Yue Z., Manish Gupta, Black Forest Labs.
Why it matters here
Controls what Arena can technically measure, including its image-generation and web/UI design arenas; any move toward expert-segmented or verified voters would come through his stack. Limited direct relevance beyond that.
How to reach
Cold. Most reachable through open-source channels (FastChat/Arena GitHub) and X. Speculation: a concrete technical contribution (e.g. an analysis of Arena's public preference data for visual tasks) is a better opener than a pitch.
What we could not establish
Undergraduate institution and PhD completion year not verified.
Few public statements of his own on data markets; views section is thin.
Miju Labs