Customers interact with an AI support assistant that can make mistakes. Which interface practice best supports human-centered transparency?

1. Prevent corrections from being submitted because disagreement makes evaluation harder.
2. Describe the assistant as infallible so users will trust every response.
3. Show only the model parameter count and omit all feedback mechanisms.
4. Explain that responses are AI-assisted, communicate relevant limitations, and provide a practical way to report errors or reach a human.