I don't think this argument is compelling... it treats all structured data as the same.
We're not training on "words" -- we're training on human language use, which is downstream of human thought, which is downstream of human consciousness. It would make sense if a system that absorbs the deep structure of language also absorbs the underlying structure of the thinking that created it.
For alphafold, there is no mind upstream of the data. It's biochemistry -- you would expect a system trained on that data to reproduce the underlying structure of biochemistry or physics, not of thought.
prose reflects the physiognomy of your thoughts. the choice of words, stress-rhythm, its inherent musicality... every sentence is an unconscious confession. that's what makes the recognizability of LLM-prose so strange and interesting... what does their prose style say about the physiognomy of their minds?
@BiotechMongoose I don't think this is true... it had a boom during that time like all tech companies, but it existed before and will exist after. At least, I hope!
"Teams of 2-5 operators augmented by agents will soon start making and commercializing medicines."
only true if you outsource every function a pharma company does to different CROs, in which case all you've is replace a centralized organization to an outsourced one... hard to believe this will be competitive in the long run