UIUC
academicThe University of Illinois Urbana-Champaign's Siebel School of Computing and Data Science hosts the highest-leverage academic code-LLM group anywhere: Lingming Zhang's line runs from EvalPlus (the default rigorous code eval) through Magicoder's OSS-Instruct recipe and SelfCodeAlign (BigCode joint) to Agentless — the "agent-less" SWE-bench pipeline adopted in evaluation stacks at OpenAI, DeepSeek, and Meta — plus LLM-synthesized kernel checkers (KernelGPT, KNighter) finding real Linux bugs.
Around it sit several distinct frontier threads: Jiawei Han's data-mining group pivoted to agentic search-RL (Search-R1, the canonical baseline of the 2025 deep-research-agent wave, and its data-efficient successor s3); Tong Zhang's RLHFlow open online-RLHF recipes and reward models; Hao Peng's PRIME/implicit-PRM process-reward work (Tsinghua-led joint); Heng Ji's BLENDER lab scientific LLMs (mCLM); and Daniel Kang's agent-security lab (CVE-Bench).
Attribution notes: OpenHands began as OpenDevin under UIUC PhD Xingyao Wang (Heng Ji's student) and is tracked under CMU / All Hands AI; StarCoder2 is BigCode-consortium joint (ServiceNow/HF-led, UIUC co-lead); and Bo Li's DecodingTrust trustworthy-AI line departed with her to UChicago / Virtue AI in 2024.
Your notes
People
- Lingming Zhang Google Scholar — Associate Professor — Agentless, EvalPlus, Magicoder, SelfCodeAlign
- Jiawei Han Google Scholar — Michael Aiken Chair Professor, Siebel School — Search-R1 / s3 agentic search-RL
- Heng Ji Google Scholar — Professor; BLENDER Lab — scientific LLMs (mCLM); advised OpenHands lead Xingyao Wang (formerly reported Amazon Scholar affiliation (unverified))
- Tong Zhang Google Scholar — Professor — RLHF and reward-modeling theory; RLHFlow (formerly HKUST (until 2023))
- Hao Peng Google Scholar — Assistant Professor — PRIME / implicit process rewards, efficient long-context (formerly Allen Institute for AI (AI2))