Summary
Wes Roth breaks down a significant internal memo from OpenAI CEO Sam Altman that reveals multiple simultaneous strategic shifts. The centerpiece is ‘Spud,’ OpenAI’s next major model that has completed pre-training — a codename that may represent GPT-5.5, GPT-6, or an entirely new product line. Altman told staff the model is strong enough to meaningfully accelerate the economy, with a release window of late March or April 2026.
The memo also confirms that Sora, OpenAI’s text-to-video model, is being phased out to free compute for Spud and a unified super app combining ChatGPT, Codex, Atlas, and OpenAI’s web tools. A previously undisclosed $1 billion Disney partnership built around Sora has reportedly collapsed as a result. The Sora research team is being redirected toward high-fidelity world simulation work for robotics — territory also being pursued by Google DeepMind and Nvidia. On the organizational side, Altman is stepping back from direct safety oversight; the safety team now reports to Mark Chen while security moves under Greg Brockman, with Altman focusing on chips, data centers, and supply chains.
The video also covers Terrence Tao’s recent paper in which the mathematician collaborated with ChatGPT and Google DeepMind’s AlphaEvolve agent to split and partially solve a complex mathematical proof — with AI carrying one of the two sub-proofs. Roth frames all of these threads as a single coherent story about OpenAI consolidating resources around one high-stakes bet.
📺 Source: Wes Roth · Published March 27, 2026
🏷️ Format: News Analysis
