Can AI make honest mistakes? (twitter.com)

"The model makes an honest mistake and mistakenly deletes $HOME instead."

AI cannot be honest or dishonest. It has no concept of what’s true or right. It’s a generative engine. We should all know that by now.

If it does something you don’t like it’s your fault for allowing it, failing to instruct the agent properly, and/or failing to watch it.

The concept of "honesty" implies a conscious agent who is making some kind of moral decision. This is anthropomorphizing plain and simple. What would constitute a "dishonest" mistake? Deliberate deception? An LLM would have to be specifically prompted in some convoluted way to do that.