AINewsnow

Normal vs uncensored: what 540 prompts across three pairs actually show

Normal vs uncensored: what 540 prompts across three pairs actually show TL;DR: We ran the same 90 prompts on three normal models and their uncensored siblings. Harmful refusal fell from 80-100 to 21-34. Helpfulness on benign trigger prompts did not improve. Capability retention was a lottery: one p…

Read the full story at DEV Community — AI ↗

Timeline · 1 report

  1. 2026-11-01 20:27 · DEV Community — AI
    Normal vs uncensored: what 540 prompts across three pairs actually show

More stories

  1. Anthropic files for blockbuster IPO as Blackstone flags likely junk rating — New York Times Technology
  2. OpenAI plans post-midterms PR push, rebranding data centers as 'AI compute campuses' — Tom's Guide AI
  3. Tech stocks slide as AI spending concerns mount — Financial Times AI
  4. Stanford HAI Fellow Raises Concerns Over AI Systems Inferring User Data Without Disclosure — Stanford HAI
  5. Saudi Aramco Acquires 49% Stake in AI Startup Humain for ~$2.6B — Techmeme
  6. DeepSeek open-sources dsh-superagent, a proactive coding agent prototype — Bloomberg AI
  7. Opinion piece urges China and US to cooperate against shared AI challenge — South China Morning Post Tech
  8. Anthropic updates Claude to stop it from recreating existing video games — Engadget

Get the daily brief of stories like this at 6:30 every morning →