Skip to Main Content

Dragon

Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

Dragon — Demo video

Wonderland: don't ask AI for the answer, ask it for a world. AI is making learning too easy. In a 2026 study, people who learned with an AI assistant scored 17 points lower on understanding than people who learned without one, because the AI did the thinking for them. Wonderland flips that. Type any "why" question, in any language, and a team of Meta models builds a small living world where the real answer is secretly at work. It takes about three minutes. Nobody tells you the answer. You predict what will happen, run experiments, put the rule in your own words, and prove it on a situation you've never seen. Meanwhile Ada, an AI scientist, races you to it. The hard part is trust, because a world that teaches the wrong science is worse than no world at all. So every world must pass four checks before anyone plays it: its science holds on every random seed, it is actually playable, every number on screen is measured and never faked, and Muse Spark 1.3 watches it run to confirm the picture matches the law. A world that fails is repaired automatically. Great interactive simulations have always taken experts months to build. PhET has made about 150 in twenty years. Wonderland builds one for any question, for about thirty cents. Teachers get a lesson guide matched to a science standard and a class link with a QR code, with no accounts needed. 28 worlds are live today, and the gallery grows every time someone asks a question. Try it: https://wonderland-omega-fawn.vercel.app

Team