Jeff Dean: belief

Transcript context

…Oh here's another interesting implication: Right now, we think about the growth in AI use as a sort of horizontal- so, suppose you're like, how many AI engineers will Google have working for it? You think about how many instances of Gemini 3 will be working at one time. If you have this, whatever you want to call it, this blob, and it can sort of organically decide how much of itself to activate, then it's more of, if you want 10 engineers worth of output, it just activates a different pattern or a larger pattern. If you want 100 engineers of output, it's not like calling more agents or more instances, it's just calling different sub-patterns. I think there's a notion of how much compute do you want to spend on this particular inference, and that should vary by factors of 10,000 for really easy things and really hard things, maybe even a million. It might be iterative, you might make a pass through the model, get some stuff, and then decide you now need to call on some other parts of the model. The other thing I would say is this sounds super complicated to deploy because it's this weird, constantly evolving thing with maybe not super optimized ways of communicating between pieces, but you can always distill from that. If you say, "This is the kind of task I really care about, let me distill from this giant kind of organic thing into something that I know can be served really efficiently," you could do that distillation process whenever you want, once a day, once an hour. That seems like it'd be kind of good. Yeah, we need better distillation.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.