12 months to the singularity
Matthew Berman showed 6 AI tools in this video on . 3 are verified available in the EU. Each row below jumps to the exact second. Part of week 2026-45.
Tools shown in this video, with timestamps
METR report, with partial internal access to Anthropic, concludes Anthropic is improving model capabilities on well-specified tasks at about twice the historical rate, with about 80% of the acceleration attributed to AI. It found no clear bottlenecks over the next 12 months, though Anthropic redacted parts and restricted the scope.
Described as the inflection point in Anthropic's internal capability index (AECI) in late 2025, when models became very good at writing code and building whole features.
Sponsor segment: runs multiple coding agents inside IntelliJ IDEA, shows session status, and sends diff comments back to the agents.
Mentioned as an earlier example of AI improving itself: Google pointed it at its hardware architecture and gained about 1% efficiency, which is large savings at Google scale.
Released early in 2026 and shown continuing the accelerating AECI curve. METR argues its jump comes from compounding AI R&D acceleration, and that unreleased internal models continue the trend.
A recent paper in which frontier AI models create new models by writing weights directly. Mentioned briefly as a sign AI may become hard to understand, with a pointer to a separate video.
