

Ah gotcha, well you hit a good point anyway as they mention the prior latest model as K2.6, which does seem like a glaring error. K2.7 has been out for a month!


Ah gotcha, well you hit a good point anyway as they mention the prior latest model as K2.6, which does seem like a glaring error. K2.7 has been out for a month!


What do you mean by this? The latest model is K3, which was released yesterday.
Edit: ah I see the reference to K2.6


It is sorta. Training is orders of magnitudes more intensive than inference, but we infer billions of times within a model generation.
A woman is eating alone at a restaurant. Her waiter makes some small talk, takes her drinks order, and brings over some nibbles. She’s friendly and the waiter thinks she’s flirting with him. She’s ready to order and thinks this is his chance, swaggers up to the table, leans over to her and whispers, “fancy a quickie?” She stands up, slaps him in the face and storms out of the restaurant. Someone from the neighbouring table says to him, “Ah, I think it’s pronounced ‘quiche.’”


I won’t speak to the purification aspect (though I suspect purification is quite trivial), but helium released into the atmosphere is wasted. Saying it’s not destroyed is by the by, we aren’t going to recover it from space as it rapidly escapes the atmosphere.
David Palmer.