Gemini's stronger reasoning lets Google expose more of the raw model

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 27:47

   puts google because google's line is probably different than some of these artists what what's your how do you think about specifically vo and flow about like how to allow artists to do crazy shit but also like the responsibility of like um not for it not to be too crazy i mean it's a great question look part of you mentioned darren uh you know he's a clear visionary right part of the reason we work started working with him early on vo is he's one of those people who's able to kind of see that future get inspired by it and kind of showing the way for how creative people can express themselves with it. Look, I think when it comes to allowing artistic free expression is one of the most important values in a society, right? I think artists have always been the ones to push boundaries, expand the frontiers of thought. And so, look, I think that's going

2. 28:52

   to be an important value we have. So I think we will provide tools and put it in the hands of artists, for them to use and put out their work. Those APIs, I mean, I almost think of that as infrastructure. Just like when you provide electricity to people or something, you want them to use it and you're not thinking about the use cases on top of it. It's a pain brush. Yeah. And so I think that's how obviously there have to be some things and society needs to decide at a fundamental level what's okay, what's not, we'll be responsible with it.

3. 29:30

   But I do think, you know, when it comes to artistic free expression, I think that's one of those values we should work hard to defend. I wonder if you can comment on maybe earlier versions of Gemini were a little bit careful on the kind of things you would be willing to answer. I just want to comment on, I was really surprised and pleasantly surprised and enjoyed the fact that Gemini 2.5 Pro is a lot less careful in a good sense.

4. 30:00

   Don't ask me why, but I've been doing a lot of research on Genghis Khan and the Esteks. So there's a lot of violence there in that history. It's a very violent history. I've also been doing a lot of research on World War I and World War II. And earlier versions of Gemini were very... Um, basically this kind of sense, are you sure you want to learn about this? And now it's actually very factual, objective, talks about very difficult parts of human history and does so with nuance and depth.

5. 30:31

   It's been really nice. But there's a line there that I guess Google has to kind of walk. I wonder if it's, and it's also an engineering challenge, how to do that at scale across all the weird queries that people ask. Can you just speak to that challenge? How do you allow Gemini to say, again, forgive, pardon my French, crazy shit, but not too crazy? I think one of the good insights here has been as the models are getting more capable, the models are really good at this stuff, right?

6. 31:04

   And so I think in some ways, maybe a year ago, the models weren't fully there. So they would also do stupid things more often. and so you know you're trying to handle those edge cases but then you make a mistake in how you handle those edge cases and it compounds but i think with 2.5 what we particularly found is once the models cross a certain level of intelligence and sophistication you know they are they are able to reason through these nuanced issues pretty well and i think users really want that right like you know you want as much access to the raw model as possible right but i think it's a great area to think about, like, you know, over time, you know, we should allow more and more closer access to it. Maybe, obviously, let people custom prompts if they wanted to, and like, you know, and, you know, experiment with it, etc. I think that's an important direction. But look,

7. 31:59

   the first principles we want to think about it is, you know, from a scientific standpoint, like making sure the models, and I'm saying scientific in the sense of like how you would approach math or physics or something like that from first principles having the models reason about the world be nuanced etc uh you know from the ground up is the right way to build these things right not like some subset of humans kind of hard coding things on top of it uh so i think it's the direction we've been taking and i think you'll see us continue to push in that direction. Yeah, I actually asked, I gave these notes, I took extensive notes, and I gave them