NEJM AI Grand Rounds AI’s Next Frontier with Dr. Kyunghyun Cho
35 snips
Mar 18, 2026 Dr. Kyunghyun Cho, NYU professor known for neural machine translation and AI in protein engineering and clinical prediction. He traces his path from language models to biology and healthcare. He explains building clinical language models, reframing AI toward operational prediction, and tools like MyChart Explorer. He stresses purpose-built data, fine-tuning, and including patients in AI governance.
AI Snips
Chapters
Books
Transcript
Episode notes
Accidental Start Into Neural Nets
- Kyunghyun Cho accidentally entered AI after being randomly assigned to a neural nets lab and learning by implementing Hopfield/Boltzmann ideas in MATLAB.
- Early struggles (looped matrix multiplies) taught him that ML spans hardware to algorithms and execution matters more than initial theory.
Open Academia Fueled Early AI Breakthroughs
- Cho highlights that early open academic research enabled rapid, shared progress because ideas were actionable and published openly before productization.
- Now product development moved into companies, making results more opaque and slowing communal knowledge transfer despite ample funding.
Language Modeling Applied To Proteins
- Cho maps language modeling to molecular biology: sequences (text or base pairs) encode higher-level 'meaning' (semantics or phenotype), so NLP techniques transfer to protein modeling.
- This conceptual parallel motivated Prescient Design work on protein function prediction and design using translation-style models.
