You will not have missed it. The text-to-video demos OpenAI put out last week have taken over every timeline I read, and for once the hype is at least pointed at something real. The clips are good. Not good-for-AI good, just good, the kind where you watch twice trying to spot the seam and don't quite manage it. After years of melting faces and six-fingered hands, that's a genuine step, and I'll say so plainly.
My actual first thought wasn't about the clips, though. It was: well, that's the end of "I saw a video of it, so it happened". Photos went a while ago, we all learned to be a bit suspicious of a still image, but moving footage still carried a residual weight of truth. A short clip felt like evidence. I don't think it gets to feel like that for much longer, and the demos this week are the moment that became obvious to everyone at once rather than just the people who follow this stuff.
I'm not doom-mongering. The same week's work that makes convincing fakes cheap also pushes hard on provenance, signing, watermarking, the boring infrastructure of "where did this actually come from". That plumbing is where I'd be spending my attention if this were my problem to solve, and it's a lot less fun to demo than a minute of photorealistic video, which is exactly why it'll lag behind. The clips are the bit that goes viral. The plumbing is the bit that matters, and it's the bit nobody's posting screenshots of this week.