@JeremiahDJohns Really depends on what you want to do with it. If you’re not coding, Claude is best. If you also want image gen, ChatGPT. If you want to do some serious dev work or research, don’t get just one - get both and compare results / have one check the other one’s work.
AIs cracking short-term superforecasting is such an important (and, still, under-discussed) trend.
Via a new piece from @nikostro https://t.co/EuVP7FyfbD
@KatjaGrace It is real. And people who insist this administration (or any) is going to negotiate productively with the CCP to get them to effectively and fully police their numerous AI labs to stay behind the US on AI development are, IMO, very naive on this matter.
@bryan_johnson “an anti-entropic system can only sustain its own order by consuming other anti-entropic systems”
Incorrect. Anti entropic systems need energy, but it doesn’t have to come from other anti entropic systems. Stars produce light that nourishes plants, stars are not anti-entropic.
@ATabarrok Reg capture locks in the current positions on the board. You don’t lock your position in when you’re losing, you lock it in when you’re winning.
@RichardHanania This is not a graph of children getting dumber. It's a graph of fewer *children being born* holding country-IQs fixed.
And it's not even in the original paper, it's other peoples' data layered into it. Don't be fooled by the quick read.
@dschwarz26 If the 22 year old was only bringing cognitive abilities that could be replaced by Fable, then sure. But maybe new hire was bringing more than that. Maybe bringing their network, their discernment, their morale-boosting, and a bunch of other things Fable can't easily reproduce.
I asked Futuresearch (top AI prediction engine) when we'd be able to fully reverse the cryopreservation of a whole mammal. The median forecast was ~40 years. https://t.co/qasfYjcm5D
@dschwarz26 I keep hearing they’re on their way to “general” intelligence, but every model I meet is frontier on just one or two narrow tasks and slow Haiku on everything else.
Put it in the interface.
1. Figure out when each question the user asked is expected to resolve by.
2. On that date, a notification appears on top of that specific question X in the side panel / history.
3. Clicking that old question brings up a quick pop: "what did you finally do regarding X?"
You can even add 50c incentive or something to the account if you want, per resolution answered.
@moreisdifferent Does it seem to provide usefully *different* answers to the same problems as Fable or Sol? Trying to decide if Grok Build would make a good secondary model, to function as a reviewer.