It’s mostly because US labs don’t care as much about it. Why waste human talent and gpus for a measly thing such as revenue when you can do super intelligence. At the rate this is going claude 6.5 might do better video than video models

Big sushi!!!!

I do not know much about the Chinese culture but in general the people seem fine and bright. The gov structure and the way it is ruled might not represent the people esp short term

@grok what percentage of Ukrainian people currently do not wanna go fight but are still conscripted, what percentage are enthusiastic and what percentage not? If there are no clear numbers available guesstimate with best available data Just deal with putin

That’s great and im sorry for your loss. The problem is someone choosing to do what they consider duty is different from taking innocents and making them fight. Are you yourself going?

Yayyy

Sure! That is the ideal ending What about the people forced to be conscripted? Placing minimum “acceptables” when you do not have to fight and avoid it like a coward while doing nothing of value is funny

Is it good in your testing?

If you weren’t a bald loser 700k kids wouldn’t have died

??? You know they will report their own numbers right?

Wasn’t 3 fucked because of some bug? Probably a way for them to get people to retry too

I mean I agree but anthropic is not worse than openai lol

True!

Tesla didn’t have the money for a compute build out

Tesla will need to solve humanoid intelligence and they will need loads of gpus for it which is why I think merger is a good idea

It’s because it is true. If the situation was reversed and tesla was trying to acquire spacex (with elon having major stake in tesla) it would still be true

You should see failure cases of v14 it is not flawless and can make deadly mistakes, they are not being overtly paranoid about it. V15 I assume will be noticeably better but still not above stupid mistakes. Will probably need ai5 and more compute and better models to avoid this

No one uses meta for llms pretty much the they use it for recommendation algos and other stuff

Yes it definitely does not have anything to do with them buying as much as they can and having ≈2.6 million h100 equivalents. Comparing to deepseek which has prolly 60k is funny