By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
No PriorsNo Priors Ep. 121 | With Chai Discovery Co-Founders Jack Dent and Joshua Meier
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
At a glance
WHAT IT’S REALLY ABOUT
AI-designed antibodies promise biotech revolution with CHAI 2.0 breakthrough
- The episode features CHAI Discovery co-founders Josh Meier and Jack Dent discussing CHAI 2.0, a zero-shot generative platform that designs therapeutic antibodies with roughly 20% hit rates, a ~100x improvement over prior computational methods.
- They explain how advances in protein structure prediction and generative modeling enable ‘Midjourney for molecules,’ allowing models to place atoms in 3D space with near-atomic accuracy to design binders for specified targets.
- Beyond faster, cheaper discovery, they argue the real impact is unlocking previously intractable targets and entirely new molecular formats, shifting biology from an empirical science toward an engineering discipline with CAD-like tooling.
- They outline CHAI’s strategy to productize the platform, open it to partners, and build interfaces that turn antibody engineers into expert prompt engineers, while scaling a small, highly interdisciplinary, engineering-centric team.
IDEAS WORTH REMEMBERING
5 ideasZero-shot generative models can radically compress antibody discovery.
CHAI 2.0 achieves ~20% success rates (hits in about 1 in 5 designs) with as few as 20 computationally designed antibodies per target, versus ~0.1% or lower for prior computational work and millions–billions of variants in traditional wet-lab screening.
Broad, realistic benchmarking is crucial to proving generalization in drug design models.
Instead of a few hand-picked targets, CHAI tested over 50 held-out targets scraped from vendor catalogs with low sequence similarity to training data, demonstrating that the model generalizes across diverse, unseen problems rather than overfitting one showcase case.
Structure prediction is the ‘ImageNet moment’; design is ‘Midjourney for molecules.’
High-accuracy structure prediction provides an atomic-level ‘microscope’ from sequence to 3D structure, while design models invert the problem by generatively placing complementary atoms to fit a given target—akin to going from classification to generative image models.
The real value is not just faster discovery, but new classes of targets and molecules.
Because AI models fail differently than traditional lab methods, they can succeed on previously intractable targets (e.g., hard-to-hit epitopes, cross-species binders, multi-target formats), expanding the reachable search space rather than only optimizing existing workflows.
Biotech is likely to see a CAD-like software stack for biology.
CHAI envisions a future where scientists specify complex design intents—multiple species, on/off-target profiles, manufacturability, stability—through rich prompts and software interfaces, with the model proposing candidate molecules much like SolidWorks or Photoshop aid engineers and designers.
WORDS WORTH SAVING
5 quotesOnce you start to grasp the implications that we are going to have the ability as a human race to engineer molecules with atomic precision, it’s almost hard to work on anything else with your life.
— Jack Dent
Structure prediction basically gives you an atomic-level microscope… once you can do that, the next question is: can we start moving those atoms around?
— Josh Meier
We’ve gone from a less than 0.1% success rate to a close to 20% success rate in a year. Who’s to say that in another year it can’t be 50% or even close to 100%?
— Jack Dent
It’s a good time to be a sick mouse.
— Josh Meier
We sort of enter this era where you have a computer-aided design suite for molecules in a way that we have SolidWorks for mechanical engineering or Photoshop for creatives.
— Jack Dent
High quality AI-generated summary created from speaker-labeled transcript.
