LLMs can't self-correct in reasoning tasks, DeepMind study finds

Re:LLMs can't self-correct in reasoning tasks, DeepMind stud... by dimgl on Monday October 09, 2023 @03:49PM (#37824622)

> I don't think I'm a fear mongerer but the amount of reasonably qualified jobs that can be displaced by the models we have now seems huge.

Maybe I'm living under a rock, but I have serious doubts about this claim.

> Think graphic designer at medium-sized law firm. Or copywriter at a scientific publisher.

How is that even possible? Take the example of the graphic designer. You'd still need them to iterate on ideas, come up with different versions for different backgrounds, generate different file types depending on the medium and format, etc. Extremely skeptical of your claims.

[ Parent | on HN ]

Re:LLMs can't self-correct in reasoning tasks, DeepMind stud... by zemvpferreira on Monday October 09, 2023 @05:37PM (#37825837)

That's the thing, a lot of professionals do rote jobs that are necessary but not valued by their employers - such as designers at law firms. Many law firms, I believe, will settle for an AI-generated version of their old logo/masthead/website and fire their designer.

[ Parent | on HN ]