Descriptions:
GPT-5.4 has officially launched, and Wes Roth breaks down its most significant capabilities against a particularly dense AI news backdrop. The model introduces native computer use—navigating desktop environments via screenshots and mouse/keyboard commands—achieving a 75% success rate on the OSWorld benchmark, surpassing both GPT-5.2’s 47% and the 72.4% human baseline. On GDPVal, which pits AI outputs against deliverables from professionals averaging 14 years of experience at firms including Deloitte, Wells Fargo, and Google, GPT-5.4 Pro achieves an 82–83% win-plus-tie rate, with an outright win rate around 70%.
The episode also covers two major industry developments. Anthropic has been formally designated a supply chain risk by the U.S. Department of Defense, applying specifically to Claude’s use within direct DoD contractor work. Anthropic has confirmed it will challenge the designation in court and has re-entered Pentagon negotiations—a development the video frames as significant but narrower in scope than initially feared. Separately, Noam Brown’s statement that AI progress faces ‘no wall’ is discussed alongside Anthropic’s newly published labor market impact report, which finds early-career hiring slowing in AI-automatable roles even as broad workforce disruption has not yet materialized.
Roth also notes that GPT-5.4 appears to borrow from Anthropic’s playbook with support for skills-based memory and migration tooling from enterprise competitors.
📺 Source: Wes Roth · Published March 06, 2026
🏷️ Format: News Analysis
