Meta's agent learns to improve itself

🔥 What happened A Meta research engineer just gave a rare inside look at his team's work on Recursive Self-Improvement (RSI): teaching AI agents to train the next generation of models themselves. He started his career fixing printers — no academic pedigree. 💡 Why it matters 90% of the job is writing code, only 10% is ideas — research engineering is data science with an infra obsession. The key insight: RSI either updates the base model or the scaffold around it. Anyone building agent harnesses today is already working on that frontier. ⚡ Our take The most honest job description we've read in months — including the confession that one bug made his results look better than they were. That kind of realism is exactly what the RSI hype threads are missing.