“You should wait to talk about your models until they are in the hands of everybody. This idea that we get a column of benchmarks that say it's 10% better than X, Y and Z has been proven again and again not to reflect practical use cases, not just for people like us doing semi-dumb stuff, but also for the high-end people.”
AI For Humans: Weekly AI News, Tools & Trends · October 2026
Context
Google had previewed a second flagship model in two weeks, but only trusted cyber-defense partners could use it. His co-host noted it beats available models on benchmarks; Gavin says he won't believe it until he can actually use it.
Transcript
don't mean anything to me until I actually get my hands. Like, yes. Yeah.
…old Argon, which was again two weeks old, down below. You know, people are saying Carbon is a clear winner. It seems to do a better job with the overall fidelity of the screen, the parallax scrolling, the graphics themselves. Hard to judge it, though, right? Hard to judge and buy some visual stuff.
No one knows. This is why like Opus 5.5 continues to impress me as how they rolled it out and what it was. Your models, you should wait until, I feel like this very strongly about this now. You should wait to talk about your models until they are in the hands of everybody. Because this idea that we get this column of benchmarks that say, hey, it's whatever, 10% better than X, Y, and Z has now proven again and again and again, not to be the practical use cases, not just for us who are doing, you know, semi-dumb stuff or things that we are kind of experimenting with, but also for the high-end people. So I really want Google to release this to the to wide so that I can use it. I might even pay for, we're going to talk about this Gemini 4, sorry, Genie 4 stuff in a bit that I might pay for the $100 model to access, but I am not going to pay $100 until I know I get access to Genie 4 and this new model. Obviously, maybe they rushed out this new AI because they were worried about the stock price thing. They were worried about having to do results. But for right now, Google, show it to me first, and then I'll…
Well, the genie is out of the bottle now. Let's talk about it. The genie's out of the bottle. From Google Deep Mind. Do I write headlines for The Verge or what? No, you really got it. They're claiming it's their most realistic and interactive world model yet. And we've covered every iteration of…
Yeah, there's a lot of really cool examples out there. There's a couple of people that were in the kind of trusted users version of this. And again, it is not rolled out wide yet, but Ethan Mollock, who has a great new book out, you should go get Ethan Mollock, was able to like get a mouse serving…
Yes. Can this AI video system be interactive game? Can it be a live streaming event that multiple people are involved with and I'm seeing the same thing interacting with it? Can it be the metaverse? Can it be the Grand Theft Auto 40? Seriously, where I'm driving the car and then I say to my buddy…
More from AI For Humans: Weekly AI News, Tools & Trends
