This AI Learned Without Humans.

Stop scrolling. An AI just learned how to train itself without a single piece of human-labeled data. And the researchers released it for free. Researchers from Tencent and Washington University created something called R-Zero. Here's the crazy part. They took one AI model and split it into two roles: ⚔️ Challenger → Creates the hardest questions possible. 🧠 Solver → Tries to solve them. No textbooks. No answer keys. No human examples. No labeled datasets. Nothing. Then they let the two AIs compete against each other. Every round: • The Challenger invents harder problems. • The Solver becomes better at reasoning. • The entire system improves itself. After just three rounds, reasoning performance jumped significantly without any human-generated training data. Why does this matter? Because today's AI companies spend millions collecting and labeling data. R-Zero suggests the next generation of AI may not need that process at all. Which means: → Faster AI progress → Lower training costs → More open-source innovation → Smarter models available to everyone The code is already public. And if this approach scales, we're watching one of the biggest shifts in AI training happen in real time. Follow for daily AI breakthroughs before everyone else starts talking about them.