Princeton
academicPrinceton's AI research runs through the Department of Computer Science and the Princeton Language and Intelligence (PLI) institute, and its defining contribution of the window is the coding-agent evaluation stack: the Narasimhan/Yang/ Jimenez/Press group's SWE-bench family (Verified, Multimodal, Multilingual) became the industry's standard frontier-model coding metric, and its SWE-agent agent-computer interface plus SWE-smith synthetic-task pipeline anchor the open coding-agent ecosystem.
Beyond agents, Princeton runs several of the field's marquee lines: Tri Dao's efficient-kernels agenda (FlashAttention-3, Mamba-2 with CMU's Albert Gu, GTA/GLA attention); Danqi Chen's princeton-nlp post-training and long-context science (SimPO, HELMET, ProLong, RLMT); Chi Jin's open theorem-proving SOTA (Goedel-Prover); and Arvind Narayanan's agent-eval rigor agenda (HAL and "AI Agents That Matter").
Industry entanglement is heavy but the entry keeps only university-shipped artifacts: Tri Dao is concurrently Chief Scientist at Together AI (co-branded on most kernel work), Narasimhan led research at Sierra AI before returning full-time in 2025 (Sierra's τ-bench is excluded), and SWE-bench Verified was a joint release with OpenAI while the dataset lives under Princeton's SWE-bench org.
People
- Danqi Chen Google Scholar — Associate Professor, CS; Associate Director, PLI; leads princeton-nlp
- Tri Dao Google Scholar — Assistant Professor, CS; Chief Scientist, Together AI
- Karthik Narasimhan Google Scholar — Associate Professor, CS; Associate Director, PLI; SWE-bench/SWE-agent PI (formerly Head of Research, Sierra AI (2023–2025))
- Arvind Narayanan Google Scholar — Professor, CS; Director, Center for Information Technology Policy (CITP)
- Chi Jin Google Scholar — Associate Professor, ECE; Goedel-Prover PI