ByteDance’s EdgeBench ran 134 tasks of twelve-plus hours each to measure what agents learn from environments rather than recall, and the noisy curves collapsed into a clean log-sigmoid law with learning speed doubling every three months.
Also: the first end-to-end agentic ransomware, and labor-force participation at a fifty-year low outside Covid — the first readout of a post-labor economy through instruments built for the old one.