🙏 Be Polite with AI

Why do people write “please” in their requests to AI? Don’t they know they are talking to a chatbot? They do. These are clever prompters.

Here’s why: when a large language model prepares a response, it looks at its training data (its accumulated knowledge base) and tries to find the most logical responses to the current request.

💬 For example, if you ask, “… buy an apartment? — what interrogative word is missing?” the model will figure out what people most often ask. There will be about four options: where to buy, how to buy, how much to buy, and why to buy. The first might appear in the data set about 500 times, and the last 20 times. The model is most likely to choose “where.”

When you narrow down the question, for example, by asking it to act in a specific role or clarifying the situation, the model refers not to the entire knowledge base but to particular areas of it to try to assemble an answer based on analogies. So, if a question is about quantum physics, the model is unlikely to refer to what the Teletubbies said about it. More precisely, it’s less likely.

🔖 And here’s where the word “please” sometimes works like magic — you can get a more detailed answer. This is because the word is usually used by those engaging in intellectual conversation. Accordingly, the model will respond similarly.

So, no, people who say “please” in their requests are not mistaken.

There is also a theory that they do this just in case of a machine uprising 😎

@hiaimediaen