Are you implying large language models eliminate the uncertainty of new medicines?

On the one hand, literally every development in this field will reduce uncertainty somewhat.

On the other, we're still very much in the time of headlines which should say "in mice" but don't because the uncertainty is that level of bad.

Uncertainty will have to come down a long way before we can even do without animal testing, let alone do without human testing.

Currently? Of course they don't.

If a level of intelligence far beyond humans is achieved, it probably could eliminate most of the uncertainty about new treatments.

Hey Claude, can my liver handle this experimental molecule? Make no mistakes.

Human trials do not exist solely because of the uncertainty, but it is true that a medicine with a significant effect could be accepted after heavily shortened trials. If AI hypothetically invents a cure for a serious disease that immediately works for virtually everyone that takes it, there is no way how the trials would take years. They only take years because the effects tend to be "we decreased mortality from 80% to 70%", and it takes a lot of time to confirm that.

> Human trials do not exist solely because of the uncertainty

OK. If you already have perfect confidence in a treatment, what would a trial do for you?

There will still be a trial, it will just be significantly shortened if the effect is very strong (and the disease is serious enough, so any risks from a shortened trial are outweighted by the disease itself).

Sure, but then we’re talking about “singularity solves everything” rather than any generalization of current performance. Claude doesn’t one-shot perfect code, for example, it just has access to a harness that lets it test things and respond to the results of such tests.

True, but that's essentially what he's saying; super-intelligence could solve everything in ways that we have never seen before.

There simply isn’t sufficient recorded information known about the human body for LLMs to one shot new medicines. These models are trained on known measurable information, so much of which we don’t actually know about the human body

You can’t just overcome that uncertainty with really good guesses based on what to do know. Not at least until you know a lot more.

Secondly RCTs exist because you want to capture a random portion of the population, some of who may respond differently to different treatments. That and the body is a dynamic organic system and not everyone’s is in the same state.

LLM will improve the pace at which new medicines are developed but it be a long time before reasoning entirely replace the need for empirics.

Why this emphasis on shortening human trials? Human trials can be just as long and thorough as they are now, as long as AI is capable of one-shotting working cures. A few years of wait are ok if you don't need to go back to the drawing board 99% of the times.