AI Just Solved Puzzles It Was Never Taught How To

NVIDIA, the company behind the chips that power most AI systems today, just announced that one of its experimental AI models, called NVIDIA AVO, solved 183 advanced puzzles across 25 different environments—with zero guidance. Think of it like handing someone a Rubik’s Cube and a chessboard, then asking them to figure out the rules and win, all at the same time. That’s what this AI did, but in seconds instead of days.

What makes this impressive isn’t just the speed—it’s the approach. Most AI today learns by being fed thousands of examples, like training a dog with treats. But this AI, like a curious child, explored and experimented to understand how things work. It didn’t need a manual because it built its own understanding of cause and effect.

So why does this matter outside of puzzles? Imagine self-driving cars that can handle unexpected situations without a programmer coding every possible scenario. Or robots in warehouses that adapt instantly when shelves are rearranged. Even your future smart home devices might learn your habits without you having to set them up.

There’s a catch, though: this isn’t ready for the real world yet. These tests were in controlled digital environments, not messy, unpredictable places like your living room or a busy street. But it’s a big step toward AI that can truly think for itself—not just follow instructions.