Descriptions:
Wes Roth covers a packed week of AI news, leading with the “Stolen Thoughts” paper — a research finding in which a team successfully extracted hidden chain-of-thought reasoning logs from model APIs at scale, raising immediate implications for AI safety, credential leakage, and model distillation by rival labs. The video explains why suppressing raw reasoning traces matters both commercially and from a safety standpoint, and why the paper’s publication is itself a significant event.
Other major stories include: Cursor’s launch of Grockbot, a persistent computer-use agent built on xAI’s Grok that functions as a virtual employee with Slack-like messaging and autonomous task execution; a reported Google AI leadership exodus; the EU passing new AI Act provisions that Roth argues are vague, outdated, and broadly harmful to compliance; Meta releasing open weights for Muse Glimmer (covered in parallel with broader community reaction); and Nvidia’s Neotron model family. Claude’s reported progress on the Riemann hypothesis — described as advancing further than any prior human effort without fully solving it — also gets brief coverage.
The video closes with a recurring segment on private-market AI valuations: Anthropic’s secondary-market price has reached approximately $1.2 trillion, up roughly 550% in a year, with Roth using this to argue that retail investors are structurally locked out of the AI boom’s most significant compounding period — framing it as context for a sponsored segment on pre-IPO equity exposure.
📺 Source: Wes Roth · Published August 12, 2026
🏷️ Format: Roundup
