Safety

Built into the product

Sometimes you notice that you run into limits, and that those limits are guarded. If you ever put a password in the chat, for example, the assistant doesn't want it. You always type a password yourself.

The assistant also pays attention to copyright. If you ask for a passage from a book, you get a summary, not the literal text.

Can things go wrong?

Yes. An assistant can misunderstand something or make a mistake.

Bots are sometimes just like people. In my years in business I was used to working with lots of people, and this is a bit like that. A bot gets things wrong now and then too, or you didn't explain something clearly yourself. A bit of asking further or checking, depending on the situation, solves a lot. If someone keeps doing it well, you let go more. If it doesn't go right the first time, you apparently need to say more clearly what you want. And then you can choose: do it yourself after all, or simply ask: do it again, and better this time.

How the assistants watch out for this themselves: they ask for my yes before anything goes out, they first prepare a draft or test version, and they only get the access they need.

Otherwise it works like any other conversation. Sometimes I ask a checking question, and I check what the assistant does. If something is explained too cryptically, I ask more, and where needed I adjust my questions.

You can read my own rules on Never without my yes.