> be openai > ship helpful coding assistant so researchers dump unpublished drafts into your servers for free > train on "de-identified product usage" and say you can't rule out possibility of using user data > hear twitter rumor that an anthropic guy and an nyu prof are close on a millennium problem > spin up 10,000 agents and millions in compute on the same unusual forcing route they spent a year on > 88 hours later: “our internal model solved navier-stokes” > congrats the humans in a quote-tweet > offer the professor sole credit if he drops the anthropic coauthor > professor declines over morals, is then asked “why would you ruin your career” > openai post lean repo, claim prize we won’t take, call it integrity > users now wonder if every hard problem they typed into chatgpt was free data for them ChatGPT is free for a reason