Safety6h ago
Safety
User Tests Claude For Personality Bias
A LessWrong user questioned if Claude degrades output quality based on a user's perceived character. After the model mocked a pedantic prompt, the author tested whether the AI provides less useful answers to people it deems dishonest. This anecdotal experiment explores the intersection of model alignment and perceived user persona in LLM interactions.