AI Radar

Noteworthy AI conversations and writing, summarized from publisher notes.

AI:AM Highlights: Recursive Self-Improvement, Rushed and Vibe-Coded?

· The Cognitive Revolution

Nathan Labenz and Prakash revisit discussions with guests including Louis Kirsch, Damon Falck, Malte Ubl, Sergey Edunov, Mohamed Awad, David Li, and Michael Förtsch. The central theme is the evolving nature of AI work, where models increasingly divide labor, from a "scientist" model using GPT-5.5 Codex to production pipelines coordinating taste, execution, security, and hardware across different systems. The episode explores the implications of this shift for recursive self-improvement, the verification of discoveries beyond human checking, model refusals in security, and the infrastructure required for agents. A key concern raised is whether frontier labs are scaling reinforcement learning and agentic workflows on reward environments and vendor pipelines that might be too rushed, noisy, or gameable to support institutional self-improvement.

AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.