News for hackers. Stuff that matters.
Re:AI is now capable of developing its own inference hardware by matteoraso on Tuesday October 06, 2026 @05:58PM (#49984704)
>It cant "invent new things". It can only reuse what it already knows about.

This is outdated. With RLVF, LLMs can create their own training data instead of relying on what it's been fed.

Re:AI is now capable of developing its own inference hardware by varispeed on Tuesday October 06, 2026 @06:00PM (#49984729)
and that training data will not be a "new thing", because AI can't invent new things. Just recycling of what it already "knows".
Re:AI is now capable of developing its own inference hardware by matteoraso on Tuesday October 06, 2026 @06:59PM (#49985386)
Yes it is. Do you not know how RLVF works? The AI tries on a task, comes up with a potential solution, and is trained based on whether the solution is valid or not. If proving a previously unproven theorem and then using that information to solve even more theorems doesn't count as creating new information, then literally everything a mathematician does is just recycling what they've been taught.
Re:AI is now capable of developing its own inference hardware by varispeed on Wednesday October 07, 2026 @08:46AM (#49992007)
Yes, but that process doesn't bring anything new, just permutation of what's been in the training data.
Re:AI is now capable of developing its own inference hardware by dwohnitmok on Tuesday October 06, 2026 @06:05PM (#49984791)
> and that training data will not be a "new thing", because AI can't invent new things. Just recycling of what it already "knows".

Regardless of the truth of the underlying assertion, this is about as textbook an example of circular logic as it gets. (at least combined with the implicit beginning assertion that "AI can't invent new things.")