Kevin Roose
speaker
381 appearances
2 recordings
1 series
first heard Oct 2025
last heard 6 Aug
Kevin Roose’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 2 in all, peaking in Aug 2026 with 1.
Appearances
They start working with like Hollywood filmmakers.
voice-verified
It can't really make what you and I think of as video, which is video with sound.
voice-verified
Yes.
voice-verified
Uh stable diffusion made models that were small enough that you could download them onto a laptop and just sort of use them without essentially any guardrails other than whatever might have been baked into the original model.
voice-verified
Fresh pasta of belly.
voice-verified
So this is Will Smith's face eating spaghetti with his bare hands.
voice-verified
His chin is kind of bobbing all around, almost like detaching from his face.
voice-verified
It's incredibly grotesque.
voice-verified
They are playing the fresh Prince of Bel Air theme song, though, which brings up warm and nostalgic feelings in me.
voice-verified
You know, I tried to get a really good answer that would be accessible to a broad audience on like why is Sora 2 better than Sora 1?
voice-verified
And the answer proved to be just dauntingly technical.
voice-verified
Like one thing that's better about Sora 2 is that it can generate videos faster than Sora 1.
voice-verified
When and so I know PJ that you're a smart guy, you're wondering like well how can it do that?
voice-verified
that?
voice-verified
The answer to that question involves step reduced solvers, rectified flow variants, and kernel level optimizations.
voice-verified
So hopefully that answers your question.
voice-verified
And I think the reason is because it's relatively easy for me to understand how a single image gets made.
voice-verified
You've probably watched this happen.
voice-verified
It uses this process called gradient diffusion, where what it's essentially trying to do is find the median in all of its training data.
voice-verified
So if you say something like, make me a picture of a bulldog on the moon, the model works and it's like we're
voice-verified
Showing 181–200 of 381 · page 10 of 20
← Previous
Next →