The idea that AI systems will soon design, code, and deploy their own successors without human help — "recursive self-improvement" — has migrated from sci-fi lore to a testable thesis, with real numbers arriving from labs on two continents.

Evidence For

ByteDance's Seed AI team published a paper this week revealing a new scaling law: AI agents double their learning speed every three months by interacting with real-world environments. The team built EdgeBench, a...