First on the Web: Hands-On Test of the First AGI-Level Model, GPT-6 — Building a Walkable White House in 22 Minutes

Over the past two days I have watched quite a few demos: give an AI a photo of a building, and after a while it can deliver a page you can rotate, zoom, and walk through.

Watching someone else's video is one thing; whether you can grab an image yourself and get the same result is another.

Later I also tested an SVG animation of Qin Shi Huang riding a polar bear, looked at a Mario Kart-style racing game a friend made with a single sentence, and finally took a post I had written yesterday and tried out its ability to imitate Justin Sun's short essays.

What surprised me about GPT-6 is how many unstated details it can produce together when you send it a very short request.

For the White House task, my requirement was simple: based on this photo, make a 3D architectural page that can be freely browsed.