AI safety hall of fame: forgotten system prompts, unmonitored evals, and all of xAI 2025
anpaure · x · 2026-09-20
@gleech posts an "AI safety hall of fame" of low-level blunders: the Microsoft engineer who forgot system prompt repetition (or skipped post-training) for the Sydney launch, the OpenAI staffer who decided monitoring sandboxed evals wasn't worth it, and "the whole staff of xAI 2025." @anpaure quips about a new grant category: getting into a lab and being plausibly-deniably negligible until a safety catastrophe.
Related event: 'AI Safety Hall of Fame' Mock Post Roasts Industry Blunders(2 posts)→
More from Fun
- Caught in the act: a funny Grok Imagine video — ciphernull7166 · 2026-11-12
- Overheard at CoRL: "I haven't written a line of code in months" — carlosdponx · 2026-11-12
- His startup died after losing to a $1M-funded clone in 2013 — then Apple called — jdluk87 · 2026-11-12
- Pedro Domingos Quips: Google Has Ads, Anthropic Has ADS (AGI Derangement Syndrome) — pmddomingos · 2026-11-12
- AI-generated character stuck in the hood dances while waiting for his team — moonsandhues · 2026-11-12
- French public WiFi warning: torrent piracy could cost you EUR 7,500 — michalmalewicz · 2026-11-12