Comment Re:AI has been oversold (Score 1) 32
Yeah. There have been productive uses of AI, like that protein folding project, but those generally are limited to very specific datasets, very specific objectives, and professionals who already know what the end goal is, they're using it as a tool and have the capability to discern quickly if what they're getting back is plausible and accurate.
But that said, there are still cases where professionals within a limited field have tried and failed to use AI. The most visible of these is the practice of law, where lawyers who don't understand how AI works used it to draft filings that were 'hallucinated' garbage full of fake citations and other things that the lawyer should have researched. I suspect that the lawyer didn't use a law-oriented AI though, just a general purpose LLM, and thus the AI was full of crap that had nothing to do with actual law. Hell, it might not have even had an index of topics specific to law and was consulting youtube comments or internet forums full of opinions that weren't based on actual law at all.
For AI to work it would have to be tailored, even siloed to specific subjects. It would need researchers maintaining its dataset. It would even need the skilled professionals using it as a timesaver to go through what it spits back at them to confirm that it's right. Recalling that law example, there's a whole lot of caselaw out there, lawyers and firms have typically had to employ armies of paralegals to do all of the research for precedent and interpretation. A law AI could help do that job if it sticks to real, actual law as-passed by legislatures and as-interpreted by courts, but even if the use of AI could reduce the labor needed to do all that research, it would still be necessary to review what the AI provided and to confirm that those citations really exist and really say what the AI thought they said.