What's this about?
A Latent Space episode with Matthew McPartlon & Neil Patil of Chai Discovery on treating drug and protein design as a scaling problem — apply more compute and data the way LLMs did, and biology's "bitter lesson" moment arrives.
What exactly is the "bitter lesson"?
A famous AI observation (Rich Sutton): over decades, general methods that just scale with compute reliably beat clever hand-crafted, domain-specific tricks. Applied to biology, it suggests big general models trained at scale will outperform carefully engineered, biology-specific pipelines for designing molecules.
Why it matters
If drug design really is "just" another scaling problem, the pace of AI-driven biology could follow the same steep curve as language models — with major implications for medicine.