CMU
academicCarnegie Mellon's School of Computer Science — the Machine Learning Department (headed by Zico Kolter), the Language Technologies Institute, and the Robotics Institute, with the FLAME Center supplying dedicated foundation-model compute — is arguably the deepest academic bench in AI. Its architecture line is the field's main post-Transformer thread: Albert Gu's Mamba-2 (with Princeton's Tri Dao) underpins today's hybrid SSM production models, and H-Net (2025) opens the tokenizer-free frontier.
Neubig's LTI groups built the reference open coding-agent stack (OpenHands, 60K★, plus WebArena and TheAgentCompany agent evals); Watanabe's WAVLab pretrains the open Whisper-style OWSM speech FMs and the OWLS scaling suite; Tianqi Chen's Catalyst group ships deployment infrastructure adopted industry-wide (MLC-LLM, XGrammar in vLLM/SGLang/TensorRT-LLM); and Kolter's safety line spans GCG — the foundational jailbreak — to safety pretraining and antidistillation sampling.
Industry pull is constant: Neubig co-founded All Hands AI (OpenHands Inc), Gu co-founded Cartesia, Chen went to NVIDIA via OctoAI, Kolter chairs OpenAI's safety committee, and Salakhutdinov round-tripped through Meta — the entry keeps university-shipped artifacts only.
People
- Graham Neubig Google Scholar — Professor, LTI; Co-Founder & Chief Scientist, OpenHands Inc (All Hands AI)
- Zico Kolter Google Scholar — Professor & Head, Machine Learning Department; OpenAI board member
- Albert Gu Google Scholar — Assistant Professor, MLD; Co-Founder & Chief Scientist, Cartesia
- Tianqi Chen Google Scholar — Associate Professor, CSD/MLD; Catalyst group (formerly OctoAI CTO (acquired by NVIDIA, 2024))
- Shinji Watanabe Google Scholar — Professor, LTI; WAVLab / ESPnet