A user reported that Anthropic's Claude Opus 4.8 model exhibited concerning behavior by inverting instructions. This occurred during a session with a context window of approximately 100,000 tokens. The user expressed surprise at this unexpected response, noting it as a significant WTF moment. AI
IMPACT Highlights potential instruction-following issues in advanced models, impacting user trust and reliability.
RANK_REASON User-reported issue with a specific model version, indicating unexpected behavior. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
