[wwweston]

> Just apply regular people filters to ChatGPT output.

This is technically correct, as LLMs are just aggregations of texts people produce.

It's also not quite right, as people have expectations/biases about how responses from computers might be different, and interaction with computers is missing many pieces of context that they might rely on to gauge probability of lying.

Telling them "ChatGPT can lie to you" is a succinct way of recalibrating expectations.