Help!
I need philanthropists to help Impetus Grants produce the missing data AI needs to ACTUALLY cure diseases of aging.
There’s a lot of talk about simulation solving biology. Useful simulations don’t come out of thin air: DeepMind’s latest weather prediction model was trained on weather records dating back to 1959. AlphaFold had decades of PDB. Tesla and Waymo’s driving simulations draw on data collected over tens of millions of miles on real roads.
Building the biological equivalent requires a much stronger data foundation. We need more observations of how diseases with long timelines like Alzheimer’s unfold and respond to treatments. Some of those observations will take years no matter how fast intelligence improves. So I’m directing @impetusgrants to identify longitudinal biosamples we can use for AI-quality data generation, as well as starting necessary new cohorts today.
I started @impetusgrants in 2021 after hearing @patrickc talk about Fast Grants at a conference. If fast grants worked for COVID, why not longevity? COVID was obviously urgent. But far more people die from age-related diseases.
Since then we’ve deployed $34M across 146 projects and supported five clinical trials, with resulting publications 100x representation in Nature/Science/Cell About 60% of the papers we assessed had no National Institute on Aging support identified. Very satisfying to see work we funded a few years ago now bearing fruit: https://t.co/AUflHFtZEA
The next round will fund around 50 projects generating open source data on how diseases of aging develop and respond to treatments. We’ve raised >$9M toward the $15M we need to kick off.
If you’d like to help fund this or know someone who could, please DM me.
(No fancy video of rockets and art deco, all our money goes to longevity research)
@JebKeiper@Polymarket Even if it happens, how do you know AI did it? I can imagine a cure for something (for example an infectious disease by eliminating a latent reservoir), but how would we define it as being AI as the primary reason?
This was a really interesting moment. When a protocol doesn’t work, my mind tends to focus on the tangible inputs
e.g. the water had the wrong minerals, the enzyme stocks were bad, the sample was degraded
It makes sense that time is a key variable. But my brain, at least, isn’t primed to care about the difference between 60 and 90 seconds
AI productivity gains in the lab might look like paying attention to the little things
Excited today to announce the first two papers from @Kimia_Tx, expanding the scope of computational methods in drug discovery on fronts both experimental (dealing with methods generating 1000x more data) and computational (speeding up search 150x):
@rbhar90 Fully agree there’s a delusion around diffusion and impact beyond CS and math. But I think we’re beyond an AI winter being possible. Pullback/slowdown yes, but AI has become too useful for the cold to really freeze progress like in the past.
The “don’t diddle your employees” rule is unbeaten.
So much value destruction avoidable by just doing the right thing.
Also “beware people telling you pleasant things that make you more money.”
Sad to see a legend like Jensen go down like this.
Since people are talking about target crowding, here's a chart from an upcoming post: <5% of new drugs entering the clinic are going after targets that haven't been tested clinically before, ~60% are going after targets with an approved drug. New ideas are hard to find!
I am going to call out large Life Science VC funds (AUM >$1B) here.
I think you have societal obligations to write sub $5m seed checks.
I know it is a pain in the ass. But just find a way to do it. All the large funds like clubby deals. Just pool 5% of your AUM into a seed check club.
I hope LPs mandate this.
@Varro_Analytics Looking at their publicly listed enterprise transactions, most are now public cos, so unsure when/what was venture at the time of writing. Could also be different financial models for the VCs. By the time it goes wrong, could be VC has already written down to 0. 🤷♂️
Today's "Opus" 5.5: I asked, "I need some ideas for how to classify chemical series in terms of their internal diversity and SAR."
Flagged.
@AnthropicAI trained on the work of all the world's scientists. It's morally wrong to gatekeep models built on their contributions.
And Opus 5.5 won’t tell me the lengths of amino acid sequences I designed with *Claude*
I have the sequences already! Do you think not telling me the lengths will prevent human extinction?!??!????????
@srikosuri hits the nail on the head:
“I’m increasingly convinced that, given the context and tools, LLMs will be as good as scientists at solving many of these problems, including the operational problems that consume so much of our time.”
My only quibble is that it’s not going to be just having a small team of scientist run a program on their own: they’ll be running multiple programs in parallel.
Full banger here: https://t.co/ykCDrKFIV1
This is the best description and implementation of how AI will improve drug discovery. I'm helping some companies implement similar processes, @OctantBio and @srikosuri are true pioneers and examples here.
Wrote an essay on how AI has been changing things at Octant, and how we've evolved from thinking about lab-in-the-loop applications to more agentic ways to work on frontier drug programs, where you don't even know what question to answer.
@zen_science I’m pretty frustrated too, but I don’t think it’s condescension. I think they deeply believe it’s risky. Probably due to a disconnect with what lab work really involves.
One question about the coming Super AI for bio: if frontier models are so good that it's "scary" why can't they distinguish "I'm trying to cure disease" from "I'm doing something bad"?
Surely that's easier than discovering entirely new biology!
@alex_morgan_x The outcome of moving the pharma industry to China isn't cheaper innovation - it's no innovation.
If America loses the political will to be the only country that pays for drugs, the biotech industry will not be profitable anywhere.