Matthew Berman: 'Hard Takeoff Has Started' — 103K Views on Recursive Self-Improvement Thesis, AI Agents Building Next-Gen AI Agents

Matthew Berman·low signal

Matthew Berman's latest video (103K views) argues we've entered the recursive self-improvement phase of AI, where agents actively help build the next version of themselves. He connects this to OpenAI's resource reallocation (killing Sora to fund Spud/next-gen models), Anthropic's Claudini autoresearch results, and the pattern of AI coding agents outperforming humans on specific engineering benchmarks. The video has sparked significant debate about whether current capability curves constitute a genuine hard takeoff.