here is another response to “well AI is good at ___” —the race to make LLMs the omniscient multitool is diverting resources and attention from the bespoke models like this protein folder, because a) most people don’t know the difference and b) this is about selling a product.
Dan McQuillanAlphaFold was always bait-and-switch; now Google confirms it. A hyper-narrow bioinformatrics model with actual science bolted on whose guesses can be experimentally checked was never a licence for socially-applied slop at scale, but it's done its job. Effort transferred to LLMs.