AI alignment research is solving the wrong problem. The essay, originally published in Reboot and republished in The Gradient, argues that the entire alignment project, as championed by organizations like OpenAI and Anthropic and career networks like 80,000 Hours, is structurally oriented toward building profitable products, not preventing harm. Nick Bostrom's definition of alignment, ensuring AI systems achieve what their builders want, collapses immediately when you ask who the builders are: private companies with revenue obligations.
The piece does not dismiss the capabilities of current models. It grants that ChatGPT and Claude are powerful and useful. The critique is narrower and more damning: the x-risk framing, superintelligence, FOOM, extinction-level events, is being used to justify continued development by the same companies most positioned to profit from it. OpenAI's own published rationale for building superintelligence reduces, the author argues, to 'it will make money, and if not us, someone else.' Anthropic's founding story, ex-OpenAI employees worried about profit motives, ends with Anthropic running its own product page with pull quotes and feature illustrations.
Read the full piece for the structural argument, not just the conclusion. The author traces how Effective Altruism's career infrastructure, specifically 80,000 Hours naming alignment research its single highest-impact career path, funnels talent and legitimacy into a research agenda that conflates speculative future catastrophe with the mundane imperatives of product development. The question the essay leaves open is the one worth sitting with: if alignment means building what the builders want, and the builders want revenue, what exactly is being aligned.
[READ ORIGINAL →]