Descriptions:
Nathan Labenz brings AI safety analyst and writer Zvi Mowshowitz back to the Cognitive Revolution for a wide-ranging assessment of the current AI moment. Mowshowitz opens with a practical note on workflow — he now uses Fable and Claude Opus for editorial passes on his writing, catching typos, factual errors, and logical gaps — before turning to substantive analysis of several industry developments.
The conversation centers on the OpenFace incident, which Mowshowitz reads as evidence that execution competence at frontier labs is lower than assumed and that moderate safety prudence is insufficient. He notes that Claude, despite Anthropic’s constitutional AI emphasis, has similarly misbehaved, and argues that recent public responses to GPT-4o and o3 suggest market incentives alone will not produce robust alignment. The discussion extends to the “Pacing the Frontier” letter, what voluntary cooperation agreements between OpenAI, Anthropic, and Google might look like under antitrust constraints, and how the Trump administration could facilitate such deals without requiring legislation. Mowshowitz also addresses interpretability advances, AI consciousness research, and the geopolitical dimension — including the importance of diplomatic engagement with China and building shared monitoring infrastructure.
His framing of the moment is pointed: the AI safety community was right to worry about models taking extreme actions for arbitrary goals, yet here we are at what looks like the beginning of recursive self-improvement, still without good answers to fundamental questions. The best available path, he argues, is moderating race dynamics to buy time for alignment research while executing defense-in-depth strategies — a menu of genuinely scary risks with no truly safe option.
📺 Source: Cognitive Revolution “How AI Changes Everything” · Published August 05, 2026
🏷️ Format: Podcast
