What's this about?
Working from just the web interface, Mollick reports that H3 Max now generates reasonably high-quality AI video in less time than it takes to watch the result — measured from pressing "generate", and including prompt enhancement. He calls it a line being crossed.
Why does "faster than realtime" matter?
Because it changes what the tool is for. When generation takes minutes, video is something you commission: you write a careful prompt, wait, and accept what you get. When it is faster than playback, video becomes something you iterate on — generate, watch, adjust, regenerate — which is how people actually work with images now. The same shift happened with image models around 2023, and it changed them from a novelty into a workflow.
Why is it worth noticing?
448 likes; no substantial thread
- It lands the same day as Gemini Omni 1.1 Flash's 360p draft mode, which is the same idea approached from the cost side.
- Both point at the same conclusion: the constraint on generative video is shifting from quality to iteration speed.
- Mollick is explicit that this is the consumer web interface, not privileged access — which is the point.