Researcher Says AI Models Remixing Literature Not Innovating
François Fleuret discusses how current AI models handle novel problems.
François Fleuret posted that state-of-the-art AI models tackle new problems by combining existing literature concepts rather than following a genuine research process such as defining minimal cases or fundamental properties. In a reply thread Lucas Beyer stated this had not been his experience in multiple cases. Fleuret then asked whether those outcomes came from fancy prompting or arose spontaneously. A separate reply agreed with Fleuret's original observation about models piping together prior ideas.
[From what I experience] even with current sota-super-smart models, brainstorming about an actually new problem consists for them to bring and try to combine existing bits from the literature, and not to do the real research process. 1/2
Combined views
Reactions from ranked influencers
5 postsNo "let's think about a minimal case", no "the fundamental properties to keep are", no "if we represent X as a Y then", etc. Just bringing tons of existing concepts and trying to pipe one in the other. 2/2
@francoisfleuret Was not the case for me in multiple cases fwiw
@giffmana Fancy prompting or it did it spontaneously?
@francoisfleuret At least this gives you a last remnant of feeling useful / needed :)