Why did OpenAI ship a dud? Ex-engineer suspects cost-cut bet on RL backfired
doodlestein · x · 2026-11-18
Engineer doodlestein analyzes OpenAI's recent poor model release: he suspects the team tried to cut inference costs with far fewer parameters, betting RL would close the gap—and that negative tester feedback was ignored. With Opus and Fable dominant and Google's Carbon strong, the timing was terrible. He diagnoses the organizational momentum: nobody inside wants to declare the model stinks and throw out months of training, especially if an impairment charge would land right before an IPO—exactly how you get a Llama 4 Maverick-scale credibility disaster. He quotes Buffett: it takes 20 years to build a reputation and five minutes to ruin it.
More from Companies & People
- No single model matches Claude, says developer as Anthropic 'hits escape velocity' — _arohan_ · 2026-11-18
- The unprecedented psychology of a scandal-hit tech CEO returning to power — signulll · 2026-11-18
- Investor: Anthropic has reached "escape velocity," lead is hard to overstate — _arohan_ · 2026-11-18
- MIT computing dean: the 'AI cuts costs' strategy is a shared hallucination — BradPorter_ · 2026-11-18
- 'Markdown sclerosis': why enterprises' stale agent instructions quietly rot AI ROI — mattbeane · 2026-11-18
- Amazon and Google to collectively own nearly 30% of Anthropic after IPO — TechCrunch AI · 2026-11-18