KAIST

KAIST is Korea's flagship science and technology university and its deepest academic AI bench, centered on the Kim Jaechul Graduate School of AI (founded 2019) and the School of Computing: Minjoon Seo's LK Lab, Sung Ju Hwang's MLAI Lab, Jaegul Choo's DAVIAN Lab, Edward Choi's medical-AI group, and Alice Oh's MARS AI Research Center. The kaist-ai HuggingFace org is largely dormant — the flagship weights ship under the prometheus-eval org.

The signature line is evaluator LMs: FLASK's skill-level evaluation led to Prometheus and Prometheus 2 — still the most-used open evaluator models — and culminated in BiGGen Bench (with LG AI Research), one of NAACL 2025's Best Papers, with a live 2026 evaluator-agent push (K-BrowseComp, peer-review benches) continuing under prometheus-eval. Transfer stories run deep: LAPA's latent-action pretraining was adopted in NVIDIA GR00T N1, and STG was upstreamed into HuggingFace Diffusers, LTX-Video, and ComfyUI.

Beyond that: Mixture-of-Recursions (NeurIPS 2025, with Google DeepMind and Mila) reset the efficiency Pareto frontier; Paper2Code (~4.8K★) turns ML papers into working repos; and Edward Choi's EHR-FM line runs from Asclepius through LabTOP (CHIL 2025 Best Paper) and the CXReasonBench / PatientSim NeurIPS 2025 D&B Spotlights. A KAIST-led consortium (HITS, Merck, AttoLab) was selected in 2025-10 for Korea's AI-Specialized FM program to build the "K-Fold" protein-complex foundation model — a watch item with no artifact yet. Startup gravity is strong: Seo founded Config (robot-data infrastructure) and Hwang founded DeepAuto.ai, whose productized artifacts are excluded here.