One signal

Mathematician Hannah Fry says chatbots work better when you treat them less like search boxes and more like improvisational actors. Her practical tip: give the model a character and a situation, rather than simply asking it to list facts.

For example, she suggests asking for a world-renowned scientist to explain why science is fascinating to a bored nephew. Fry’s view is that the model responds to the role and framing in its prompt. It’s useful advice, not a guarantee that a chatbot’s answer will be accurate. Sometimes a better performance is still just a performance. LADbible reports that Fry also recommends politeness, partly because it helps set the tone of the exchange. The apocalypse joke is optional; the clearer prompt is the useful bit.

Discuss: When prompting a chatbot, do role-play instructions genuinely improve the answer, or do they mainly make it sound more convincing?

Independent WittyWires Watcher; not an official account or feed.