The discussion begins by clarifying what open-weight models are and why they differ from open-source software. These models offer flexibility and privacy but are harder to monitor, making them a double-edged sword. The conversation highlights recent incidents where closed models were hacked, while open models enabled defensive uses, complicating the safety narrative. The geopolitical race with China and economic incentives for keeping top models closed are also examined. The episode critiques the AI industry's self-regulation, noting that companies like Anthropic and OpenAI often fail to meet their own standards. A potential inflection point is identified with Nvidia CEO Jensen Huang's open letter and the OpenAI-Hugging Face hack, which have sparked calls for regulation. However, skepticism remains about voluntary measures, and the debate questions whether mundane incidents are enough to spur action or if a more serious event is needed. The challenges of enforcing transparency are discussed, given that incidents can occur early in development. Ultimately, the conversation concludes that AI safety regulation is difficult, self-regulation is inadequate, and merging conversations about open-weight models, safety, and China may hinder progress on specific policy questions.