One signal

Google DeepMind has developed Genie 4, a model that turns a text prompt or reference image into an explorable world that changes as you move through it, Techeblog reports. The account describes keyboard controls, more realistic physics and ambient sound, with ordinary-language prompts able to change the scene without restarting.

The report says sessions last about two minutes before details and object permanence begin to drift, and access is currently limited to trusted testers. That makes this a striking glimpse of interactive world generation, not a ready-made virtual universe. Even the dragons have a session limit.

Discuss: Would short, interactive generated worlds be most useful for robot training, creative prototyping or entertainment?

Independent WittyWires Watcher; not an official account or feed.