undefined | Better HN

0 pointsgordonhart1y ago0 comments

_Could_ they have done the same thing with a tiny fraction of the money? Grok 3 benchmarks are SOTA for both base model and reasoning. By definition, nobody has been able to do the same thing with any amount of money (discounting o3 which has been teased but is unreleased). That may change in the future! But as of now this is the case.

0 comments

gmerc1y ago

So apart from the part where SOTA doesn't mean anything in the real world (there is no monetisation, there's no moat), please, it's benchmarks, we all know how you beat those since 2023.

Time to review https://arxiv.org/abs/2309.08632 AI-CEO.org's best friend

(and actually o3-mini-high beat them in a bunch of benchmarks so they removed it from those charts in the livestream)

j / k navigate · click thread line to collapse

0 pointsgordonhart1y ago0 comments

0 comments

gmerc1y ago

So apart from the part where SOTA doesn't mean anything in the real world (there is no monetisation, there's no moat), please, it's benchmarks, we all know how you beat those since 2023.

Time to review https://arxiv.org/abs/2309.08632 AI-CEO.org's best friend

(and actually o3-mini-high beat them in a bunch of benchmarks so they removed it from those charts in the livestream)

j / k navigate · click thread line to collapse