KM4BI

GPT-6 Astra Saturates ARC-AGI-3, Tesla's Cybercab Reaches Austin, Anthropic Formalizes Fermat's Last Theorem

Frontier AI, robotaxis and formal proof: three signals of an accelerating technology cycle reshaping cost structures and capabilities.

Published

This episode connects three announcements that point to accelerating technical capability: GPT-6 Astra, Tesla Cybercabs appearing in Austin, and Anthropic’s formalization of Fermat’s Last Theorem.

Models designed to act

The panel describes Astra as a frontier model aimed at computer use: understanding screens, images and interfaces, then completing work in a tight interactive loop. The cited benchmark results are striking, but the discussion also stresses that scores need to be tested against real workflows, especially when a model operates in software environments.

Robotaxis as infrastructure

The Cybercab is presented as a two-seat vehicle with no steering wheel or pedals, designed around autonomous driving. The economic case rests on hardware simplification: fewer components, an approximately $30,000 target price mentioned in the conversation, and the prospect of owner-operated fleets. Local approvals, safety and manufacturing execution remain decisive.

Formal proof at a new scale

The conversation cites a formalization of Fermat’s Last Theorem in roughly 13 million lines of code, proving about 29,000 theorems along the way. The point is not just the result: it is the prospect of turning large-scale reasoning into verifiable artifacts that others can build on.

What to watch

Competition between AI labs is increasingly about agents, efficiency and tool integration. At the same time, autonomous transport is moving from demonstration to operating economics, while formal verification is becoming a concrete marker of progress in scientific AI.

Source