OpenAI LLM flags ML paper title 'Deep Ignorance' as biological risk, refuses task
burkov · x · 2026-10-21
Andriy Burkov found OpenAI's LLM refused to extract information from a paper titled "Deep Ignorance: Filtering Pretraining Data Builds Tamper-Resistant Safeguards into Open-Weight LLMs," flagging it for possible biological risk. A screenshot shows an HTTP 400 error advising rephrasing, linking to OpenAI's biology-safety blog post. The irony: a paper about tamper-resistant safeguards got blocked by overzealous safeguards — Deep Ignorance in its glory.
More from Fun
- Dev adds 'feel free to scroll TikTok for a few hours, you earned it' to agent prompts — jaivinwylde · 2026-10-21
- Beff Jezos reacts to 'The Beaff' meme: someone call the Halloween store — beffjezos · 2026-10-21
- GitHub bug deleted YC startup's entire org with 10k-star repo; recovery fumbled for weeks — kingslyj · 2026-10-21
- Longtime Suno user roasts its vocal shimmer artifact: 'makes me puke' — lfschiavo · 2026-10-21
- Hangzhou: China's greenest city and home to Qwen, DeepSeek, and Unitree — vincent_koc · 2026-10-21
- Google scientist visits humanoid robot fight event, poses with suspended robot — edgarpavlovsky · 2026-10-21