The Great Pretender

Large language models will give you better answers if they are clear about what role they should take on in a conversation. This can be incredibly useful, especially in ongoing conversations.

It’s possible that the first demos you’ve seen of ChatGPT, or have tried yourself, were about the AI answering in the voice of a pirate:

What is AI? Explain it like a pirate.

The answer:

Ahoy matey! AI be short fer "artificial intelligence." It be like havin' a crew o' robots that can think an' learn like humans do. They be programmed to do tasks an' make decisions on their own, without needin' a human to tell 'em what to do all the time. It be a powerful tool fer navigatin' the high seas o' technology, and it be growin' stronger every day. Arrr!

This is both funny and very useful.

You’ll notice that this isn’t a just simple translation of a standard answer into pirate lingo. Instead, the LLM uses metaphors that make sense in the context of its role as a pirate, like navigating the high seas of technology.

Telling the LLM what role it should take on isn’t just good for laughs. It’s an incredibly powerful method to get better answers. Let’s take a deeper look.