logoalt Hacker News

michael0churchyesterday at 10:19 PM1 replyview on HN

Right. And I bet this was to some degree accidental; that no one ever intended to absorb unpublished work into training data; it just happened. AI is still mediocre (often, bad and getting worse as novels benchmark) when it comes to novelty and creativity, but it can reason within what it has memorized quite well.

The part that makes them horrible that can’t be charged off as an accident is the insistence that an academic scrub someone off a math paper because he worked at Anthropic.


Replies

wieitoday at 12:28 AM

Was it?

They've been dishing out cheap access specifically to researchers give over lmao. The researcher's got lured in - they need to accept they got played TBH.

Altman is certainly more devious than Amodei - he's shown that time and time again.

PG was right about he said about him.

Every entity on earth should see it as a kill shot: be careful what you put in the models. None of your information is safe.