They don't necessarily even hint them. I think the car mainly asks them yes/no questions and they respond.
16,618 karma · joined August 28, 2009
It doesn't even prove the models do that. The RLVR environments being mostly Python isn't "training data memorization". That's just the kind of dumb thing people say to sound savvy.
True, but just don't do that then.
If you send me the number it will get looked at.
Or for general memory `footprint` is good.
Those investors don't want to give /you/ money, they want to give the billionaire money.
That's only for spatial audio.
(But it's encrypted, so you'd better have backups because you can't read it off the chips.)
> Windows/Linux treat RAM like a stream of bytes (similar to how you’d compress a .zip file) so it’s not as efficient.
That doesn't really follow. There are faster and slower compression algorithms no matter what, and 64-bit integers are kind of a waste of memory much of the time.
Also, unified memory has tradeoffs. The GPU improvements are real but it mostly means more pressured on memory, not less.
The plan mode is useful because if you do corrections during development mode it does that silly thing where it leaves comments referring to your corrections.
Pangram is pretty reliable and shows 100% AI.
(Also, political donors are unusually engaged upper-middle-class people which means they are left of the average voter.)
https://www.anthropic.com/constitution
> Anthropic wants Claude to be genuinely helpful to the people it works with or on behalf of, as well as to society, while avoiding actions that are unsafe, unethical, or deceptive.
The 5.x series have terrible writing styles, which is one way to cut down on sycophancy.