Chrisprattalpharaptr
Ave Imperaptor
No bio...
User ID: 1864
Chrisprattalpharaptr
Ave Imperaptor
12d ago
The promise of AI from the beginning has been 'Cure cancer (and all other diseases).' Oh, and immortality too.
How was that supposed to happen without a wet lab? I guess you could try your hand at improving existing in silico simulations, but even that would take a lot of wet lab benchmarking and at the moment SOTA is nowhere near sufficient along several axes.
That being said, I've chatted with people who have an idea of the projects they're going after and the ambition is several OOM less ambitious than curing cancer and pretty far from GoF research. Dario specifically cites rare diseases in the article so it's probably in the vein of 'Millennium problem, but for biology' so they can get press along the lines of 'ANTHROPIC DISCOVERS NEW DISEASE AND CURES PATIENTS.' Which would hype investors and e/acc types, but be meaningless as far as having any kind of impact on people's lives.
5 years from now? Who knows. Maybe I'll at least learn the definitive answers to all the questions I studied in grad school before I get turbo-COVID.
Chrisprattalpharaptr
Ave Imperaptor
17d ago
They do reward hack etc of course, and this is very concerning, but even with e.g. the HF hack they showed no interest in figuring out how they could convert the universe to computronium so they could receive eternal blessings from the Grader.
Honestly I was a bit confused on this point. In the message logs you see them frequently mentioning that their compute budget is almost up - do none of them hit on the possibility of just hacking the server to give themselves infinite compute budget? Or do they come to that realization and just fail to procure more resources?
Chrisprattalpharaptr
Ave Imperaptor
17d ago
It's possible that there'll be an AI improvement plateau in between "superhuman mathematician-hackers" and e.g. "superhuman industrial biowarfare experts"... Either way, every year of delay means an extra 10M cancer deaths worldwide.
I can guarantee you that they could easily be superhuman biowarfare experts if they wanted to. The unknowns would be whether they could find a way to produce their viruses without being detected.
Curing cancer is a completely different OOM of difficulty. A human intelligence could design a good-enough virus just by mixing and matching existing parts. 'Curing cancer' or aging would take entirely new paradigms or modalities and probably a willingness to modify our own biology pretty significantly. Sticking with existing paradigms gives you pretty drastically diminishing returns; there's only 20,000 odd genes for AI to make a drug against, and I'd wager that a lot of the best ones have been found and this is a large part of Eroom's law.
Chrisprattalpharaptr
Ave Imperaptor
18d ago
Other sub cultures and scenes have been destroyed by the internet and modernity.
They're still around if you know where to look. I found plenty adjacent to bike culture (NSFW for nudity), and most large cities in the USA have several annual bike rides/budget bike workshops with afterparties and social scenes. Swing dancing, the performing and visual art scenes, cottagecore groups and community gardens, certain pickup sports all had vibrant social scenes. Most of them just have people that (no offense) you and others here probably detest and wouldn't want to hang out with.
Starting with the mouse/tweet from Patrick Heizer first:
Mouse models in cancer are a very contrived system (with varying levels of contrivedness). Human tumors evolve over months/years while exposed to selective pressures from the immune system and other defenses against cancerous cells. They recruit immunosuppressive cells, their own blood supply and dense fibrous tissue to shield themselves from the rest of the body and ultimately metastasize. They're extremely genetically heterogeneous. And the mice have robust immune systems and are otherwise healthy, whereas human cancer patients skew old, they're often blasted with chemo or otherwise sick, etc. I'm not a cancer biologist, but I'm sure an LLM could do a better job listing all the differences.
The most basic murine tumor models just involve injecting them with a single genetically homogenous cancer cell line which lacks all of the features mentioned above. Trivially easy to clear, and most of the reddit-tier hype articles are just showing 3rd tier journals with results of this type.
Genetic murine tumor models and patient-derived xenografts are both big improvements, but still have various deficiencies compared to the above. And even in these models, we often advance clinical candidates that only delay death in the mice.
Lastly, the types of cancers you can target with immunotherapy (your mRNA vaccines) are limited. There's a spectrum of # of mutations in a given cancer, and melanoma is at the high end given that it's usually caused by UV irradiation. That's why the initial PD-1 and more recent Moderna data were both done in melanoma.
That's what we'd call and objective response. And that's great! But traditional chemos also have pretty robust objective response rates. The main readouts in most of these cancer trials is progression-free survival (PFS) or overall survival (OS). Lots of therapies can generate ORs; the vast majority are not cures.
The other thing worth keeping in mind is that these personalized vaccines are going to cost a bajillion dollars. I guarantee that when Moderna prices their cancer vaccine and it hits the news, we're going to get a spate of news articles hyping it and pharma squeezing desperate sick patients. CAR-T therapy costs 0.5-1 million dollars, and companies are still losing money doing it. There's probably an argument for relaxing stringency of the manufactured product (a large part of why it's so expensive), but relaxing regulation on big pharma is generally a losing argument to the public.
Manufacturing the mRNA vaccines won't be as bad, but making a single dose is going to involve the clinicians taking a biopsy, running the clinical-grade sequencing, paying the bioinformaticians to search for epitopes, some time-sensitive safety testing run by clinical-grade lab grunts and probably half a dozen other things I'm forgetting.
A cure for cancer is going to involve some new treatment modality or paradigm. I'm interested to see what the swarm comes up with.
More options
Context Copy link