Chai Discovery's Bitter Lesson: Drug Design Is Another Scaling Problem

Summary

The video is a Sequoia-hosted interview with Chai Discovery co-founders Josh Meier and Matt McPartlon about applying AI scaling laws to drug discovery. They describe how diffusion models and large protein datasets enable de novo antibody and molecule design, with Chai-2 lifting hit rates from sub-0.1% to roughly 15-16%. The company's strategy is to build a computer-aided molecular design suite and partner with large pharma rather than develop its own drug pipeline. No public-market trade or investable security is explicitly recommended; Chai Discovery itself is private.

• Chai Discovery applies AI scaling laws to protein and drug design.
• The field evolved from protein folding to generative diffusion models for protein design.
• Chai-2 raised antibody design hit rates from below 0.1% to about 15-16%.
• The company aims to build a computer-aided molecular design suite and shorten design loops.
• Its business model partners with pharma instead of building a full drug pipeline.
• Data flywheels use the Protein Data Bank, sequence databases, and generated lab data.
• The founders discuss talent, culture, compute scale, and competitive pressure in AI drug discovery.
• No public security or investable instrument is explicitly recommended in the transcript.