X users roast OpenAI's obscene profits from an 'accidental' transformer discovery
akbirthko · x · 2026-11-10
LinusMixson posted that OpenAI (and to a lesser extent Anthropic) stand to profit "obscenely (and, in the minds of many, eternally)" from a basically accidental discovery: overtraining a transformer does something weird. akbirthko reposted with "there are no accidents." The attached image shows an agent pulling page 53 of Ilya Sutskever's 2013 PhD thesis, which speculates that achieving the asymptotic limit in text compression requires understanding "equivalent to intelligence."
More from AGI Musings
- Matt Pocock: agents supercharge Goodhart's Law — PR-count metrics now actively harmful — mattpocockuk · 2026-11-10
- Norway runs 2,000+ underwater AI laser robots delousing 50 million salmon — jjvincent · 2026-11-10
- Stanford's free CS193T 'Thinking with AI' course needs zero technical background — JeremyNguyenPhD · 2026-11-10
- PE investor refuses hourly-billed AI consultancies: 'This job costs $5,000 — whether it takes 5 hours or 2, we don't care' — alex_verem · 2026-11-10
- Labs are keeping their most capable cyber, math and bio models internal, warns lvwerra — TheZachMueller · 2026-11-10
- In the age of vertiginous AI ascent, the stiff-drink-first movie trope makes more sense — StewartalsopIII · 2026-11-10