Commenters read OpenAI’s embedded assessments as an effort to scrutinize AI safety practices, but disagree over whether the approach is useful or overly constrained. One commenter welcomes it as hard to criticize, while another calls it a guarded, “hobbled” external audit. A reply argues that assessors get more access to internal tools and resources than a conventional audit would allow. The term “embedded” also draws debate: one commenter prefers “on-site” to avoid military associations, while another says “embedded” is ordinary language. A joke recasts the role as an expensive consultant sent to sit in a cubicle.
The initiative prompts broader speculation about OpenAI’s direction relative to Anthropic. One commenter sees Anthropic advancing while OpenAI slows for safety, and weighs differences in interpretability, alignment, and IPO pressures; they would currently choose to lead Anthropic, though they think OpenAI could regain the lead. Another argues OpenAI has become driven by fear of public backlash and risks losing safety talent to Anthropic, while lamenting that both companies may assume authority over alignment. A separate commenter suggests liability, rather than good intentions, may explain differences between the firms. Others focus less on safety strategy: one asks when a better model will arrive and complains about the price of the current product.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.