GPT-5.6 reactions, Phoebe Gates' Phia backlash, and Hollywood's chase for internet horror

Diet TBPN connects GPT-5.6, SK Hynix's US listing, Meta's AI thesis, Phia's affiliate controversy, and Hollywood's hunt for internet-native horror IP.

This Diet TBPN episode links several signals about how AI is reshaping markets, software interfaces, and internet culture. It begins with SK Hynix's US ADR debut, framed as a way for investors to gain cleaner exposure to high-bandwidth memory, one of the key bottlenecks behind advanced AI systems.

The hosts then move to GPT-5.6 demos: 3D scenes, Blender-style workflows, web design, logos, and computer-use examples where the model appears to operate software at a speed no human workflow could match. The point is not just that the demos are impressive; it is that the creative layer still matters. Recreating an existing film is less valuable than using the new tools to make something original.

OpenAI's desktop app shift becomes a product-design lesson. A unified app that brings ChatGPT Work and Codex closer together may be strategically logical, but it also disrupts users who relied on the older native Mac app, familiar shortcuts, and established workflows. The episode treats that friction as a reminder that model capability and interface continuity are separate problems.

The Meta segment discusses a SemiAnalysis argument: catching up in frontier AI requires models, data, talent, and compute. The most interesting piece is Meta's ability to redirect engineers toward reinforcement-learning environments, suggesting that high-quality training data is becoming a strategic engineering discipline rather than low-end labeling work.

The closing sections focus on attribution and intellectual property. Phia, the shopping app co-founded by Phoebe Gates, is discussed through allegations that its extension may inject affiliate codes in the background without explicit user interaction. Hollywood, meanwhile, is buying up internet-native horror properties such as Siren Head and Mandela Catalogue, treating YouTube, Reddit, and Roblox as discovery engines for film concepts with proven demand.

What to watch

Source