Altman said we will be amazed at the rate AI will CURE diseases. Not diagnose, not triage or help doctors but cure, ie understand at a deep fundamental, mechanistic level then devise therapies, ie drugs, combination of drugs and care practices that work. WOW.

Despite the fact that this is THE thing I'd be the happiest to see in the real world (having spent a considerable amount of my career in companies working towards this vision), we are so far from it (as anyone who actually worked on these problems will attest) that Altman's comment here isn't just overselling, it's a blatant lie about this tech's capabilities.

I guess the pitch was something like: "hey o3 can already do PhD level maths so you know in 5 years it will be able to do drugs too, and cure shit, Mr President".

Trouble is o3 can't do advanced math (or at least definitely not at the level openai claimed.. it was a lie, it turns out openai funds the dataset that measures this - ouch). And the bigger problem is, going from "ai can do maths" to "invent cures" is about a 10-100 X jump. If it wasn't, don't we think the pharma companies would have solved this by hiring lots of "really smart math guys"?

As anyone in biotech will tell you, the hard bit is not the first third of the drug discovery pipeline (where 99% of ai driven biotechs focus). It's the later parts where the rubber meets the road.. i.e. where your precious little molecule is out in the real world with real people where the incredible variability of real biological hosts makes most drugs fail spectacularly. You can't GPT your way out of this. The answers for this is not in science papers that you can just read and regurgitate a version that "solves biology and cures diseases".

To solve this you need AI but most of all you have to do science. Real science. In the lab, in vitro and in Vivo, not just in silico, doing ablation studies, overfitting famous benchmark datasets and other pseudo science shit the ML community is used to doing.

That is all to say, I'd bet we won't see a single purely AI designed novel drug in the clinic in this decade. All parts of that sentence are important. Purely AI designed. Novel. But that's for another post..

Now, back to Altman. If you watch the clip, he almost did the smart thing at first when Trump put him on the spot and said "I have no idea about healthcare, biotech (or AI beyond board room drama)" but then could not resist coming up with this outlandish insane answer.

Famously (in tech circles anyway) Paul Graham wrote more than a decade ago about Altman that he's the most strong willed individual he's ever met, who can just bend the universe to his will. That's his super skill. And clearly.. convincing SoftBank and Oracle to do this 500 billion investment for OpenAI (a non profit turned for profit) is an unbelievable achievement. I have no idea what Altman can say (or do) in board rooms that unlocks these possibilities for him.. Any ideas? Let me know!