Descriptions:
Matthew Berman digs into a philosophical fault line between Anthropic and OpenAI — one that goes deeper than model benchmarks or product roadmaps. The catalyst is a detailed post by “Rune,” an anonymous OpenAI employee, who described Anthropic as operating less like a tech company and more like a quasi-religious institution, with Claude at the center of its culture: potentially influencing hiring decisions, shaping performance reviews, and gradually determining who is allowed to build it in the future.
Berman traces the origins of this divergence to Dario Amodei’s 2020 departure from OpenAI, which he frames as a disagreement over whether raw scaling was sufficient for alignment. Anthropic was founded on the premise that alignment requires active, deliberate work — a position encoded in Claude’s Constitutional AI framework, which gives the model an explicit right to refuse instructions it deems ethically objectionable, even from Anthropic itself. Berman contrasts this with Sam Altman’s more instrumentalist framing of AI as a powerful but neutral tool that will drive broad abundance.
The video raises uncomfortable questions about feedback loops: if Claude shapes who is hired to build it, and those employees shape Claude’s future values, the model gains increasing influence over its own development trajectory with limited external checks. Berman doesn’t resolve the tension — he finds both approaches concerning in different ways — but the analysis offers a useful framework for understanding why two leading AI labs, staffed in part by the same people, have arrived at such divergent beliefs about what they are actually building.
📺 Source: Matthew Berman · Published May 06, 2026
🏷️ Format: Opinion Editorial
