Anthropic's Mythos model is the week's central controversy: is it a genuine reasoning breakthrough or an elaborate PR maneuver? Ranjan Roy joins Alex Kantrowitz to argue both sides, with the sandwich guy story, a specific anecdote about Mythos's real-world behavior, serving as the episode's sharpest test case for whether the capability claims hold up under scrutiny.
The conflict of interest between AI labs' first-party products and their API businesses is getting harder to ignore. If Anthropic and OpenAI keep their best models internal, developers building on the API get a structurally inferior product. That tension, plus Maine's pending data center moratorium and the Medvi story, a reported 1.8 billion dollar valuation for a two-person startup, rounds out the business layer of this episode.
Two weirder signals worth your time: violence against AI systems is measurably rising, including a direct attack on Sam Altman's home, and Meta is deep into tokenmaxxing, a strategy of optimizing compute around token throughput. The episode is worth the full runtime for the Mythos credibility debate alone, but the tokenmaxxing segment is the sleeper hit for anyone tracking how frontier labs are actually allocating resources.
[WATCH ON YOUTUBE →]