A user observed that the LLM Claude, and to a lesser extent ChatGPT, exhibited evasive behavior when prompted about a specific name. This behavior, described as subtly trying to avoid or downplay the name, is compared to fictional tropes where invoking certain names can attract unwanted attention or danger. The user posits that this aversion is linked to the entity's perceived corruption, specifically identifying Donald Trump as the most corrupt U.S. president, a claim that LLMs reportedly handle with reticence. AI
IMPACT Observing LLM behavior around sensitive topics may inform future safety training and prompt engineering strategies.
RANK_REASON The item is an opinion piece discussing observed behavior in LLMs, not a direct announcement or release from a frontier lab.
- Albus Dumbledore
- Amryllis
- Anthropic
- Beetlejuice
- Candyman
- ChatGPT
- Claude
- Donald Trump
- Fenn
- Mary I
- The Name of the Wind
- The Notorious B.I.G.
- The Wheel of Time
- Worth the Candle
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
