And all of those things are then analyzed and verified before anything is done with them. No reputable scientist is taking those results and dumping it straight into a paper; the deep learning engines are pointing scientists in the right direction; they’re taking the haystack and making it a handful. Protein folding is a little different because the results can be directly verified programmatically (I think; I’m not an organic chemist, or biologist, or whoever is doing this research).

The output of LLMs can be great outlines. They can also be wildly, and confidently, wrong.