Latent Space
🔬 The Coolest Diffusion Research Isn't in LLMs — Evan Feinberg & Sergey Edunov, Genesis Molecular AI
agentic
What happened
Latent Space interviews the founders of Genesis Molecular AI, including the former Llama lead from Meta. They discuss their flagship model, PEARL, which achieves high accuracy in small molecule drug discovery and co-folding. Crucially, they highlight how these highly accurate physical/molecular models enable entirely new agentic workflows for automated drug discovery.
Why it matters
Highly accurate domain-specific models are unlocking autonomous agentic workflows in complex scientific fields like drug discovery.
The take
While biotech is niche, the pattern of 'high-accuracy domain model + agentic workflow' is a blueprint for other physical and complex engineering industries. It shows that agents are only as good as the underlying predictive models they use as tools.
Do this
Watch the interview to understand how domain-specific accuracy thresholds enable agentic automation in scientific computing.
Don't read this site daily. Get it in your inbox.
The daily brief and Sunday deep dive — distilled, scored, and opinionated. For builders only.