Sounds like a great name for a conference 👀

00:53 👀👀👀

Look ma, it me!

What made it special was your request for speakers to share something new. Means it wasn’t just regurgitated slide decks and marketing pitches.

UNLOCK2026 was something like a labor of love for the AIxScience community. One founder told me it was the best event he’s attended since starting his company 8 yrs ago. Another said it felt like JPM 2.0 🤯 If you missed it, we’ll be sharing videos next week. Stay tuned.

Spent the last week in SF learning about AI x bio: 🧵 1. The vast majority of science is done on <100 organisms. Transferring some intervention you came up for e. coli to lactobacillus is incredibly hard, even though both are common bacteria. Think about how hard it is to port LLMs across accelerators even though the primitives are the same. Similarly, even when the genetic code is almost the same, the way in which those genes are expressed can be very different, the production of some protein may require cofactors that the second organism doesn't produce, etc. This problem is so hard that most people don't even bother trying to address it (s/o to the folks at Cultivarium for being one of the few and explaining this to me in detail). 2. AI will lead to pipeline abundance but maybe not outcome abundance. The two big bottlenecks are financing and regulation; for example, just because you can come up with more drugs does not mean you can test all of them in the real world, nor is it financially viable to do so. For these reasons, there probably won't be a ChatGPT moment in AI x bio. There's a small chance that regulation also changes to keep up with the tech, but industry folks aren't optimistic. 3. All the frontier labs are making a push into the bio space, some more so than others (see: Anthropic's recent acquisition of Coefficient Bio). Interestingly, I didn't get the sense that the latter was building its own foundation models. According to a cofounder, they seemed to be focused on context engineering and creating tools with good "ergonomics" for frontier models. This particular acquisition was even more surprising given that frontier labs typically partner with several different startups rather than picking winners and losers early on. 4. Biz dev is a huge problem. Big pharma companies are used to outsourcing risk and innovation to startups, but the terms used to be clear: come up with useful assets (e.g., a working drug), and we'll license it / buy it / buy you. It's not clear how deals should work with AI companies --- do they get a percentage of eventual product revenue? An ongoing subscription fee for using the tech? Equity investments and tight partnerships? There's no real template right now and most deals are being constructed as they go along. Common sentiment that tech transfer and systems integration is broken. Many thanks to @michellearning and @medra_ai for hosting the Unlock event --- I learned a lot!

Welcome to the scientific revolution. 100s of robots. Zero coffee breaks. America’s largest autonomous lab, open today.

When the rockstar of biotech talks, we listen! Thank you Aviv Regev of @genentech for capping off an amazing day of talks, panels, and discussion. #UNLOCK26

Only pharma embracing AI will survive the next wave - @AstraZeneca’s Chief of AI for Science Innovation Jorge Reis-Filho is telling us how they’re going to thrive - only at UNLOCK. Moderator: @michellearning @medra_ai #UNLOCK26

Next panel on deck ⚾️ @EliLillyandCo, @arcinstitute , @Ginkgo, and @newlimit ready to go

Proudly nerdy 🤓

Standing room only 👀 Tough questions from @chrisleiter_, better answers from @JCoolScience (@AnthropicAI ), Sarah Nam (@abbvie), Eli Berlin (@Terray_Tx), and Stacy Feld (@JNJInnovation) on which partnerships are actually happening and working. #UNLOCK26