Keep on to blur preview images; turn off to show them clearly
Every problem we imagine to test intelligence, can also be solved without intelligence. So we must inspect not just whether a problem is solved, but how quickly, how cheaply, with what prior knowledge, and whether the way it was solved transfers well to other problems.
Sculpting, AI, Philosophy, Coding
Hmm. Or maybe I'm overthinking it and we just use what works to construct something more robust from lots of LLMs.
Sculpting, AI, Philosophy, Coding
NanBanPro image of what I was saying in January. Do AI labs investigate bottom-up training for LLMs much? I don't think embodiment is necessary for intelligence, but it sure would be handy for bottom-up learning... But that's way harder to scale, even if it's a virtual body.
Sculpting, AI, Philosophy, Coding
Still one of the more interesting plots from Artificial Analysis. Makes me pretty intrigued by Grok 4.1 Fast. Here's to more efficiency gains over the next year, so we can get more dots in the top of that green zone!
Sculpting, AI, Philosophy, Coding
Nano Banana Pro still can't count particularly well... (with the classic @ESYudkowsky prompt)
Sculpting, AI, Philosophy, Coding
This surprised me, given that the translation result without the custom instructions didn't even have em-dashes. And by 'borderline wrong', I mean that the translation itself was technically correct, but didn't fit the mood of the original message at all, sounded very depressed.
Sculpting, AI, Philosophy, Coding