Telling ChatGPT to “sound more human’” misses the point

When you ask ChatGPT a question, it doesn’t even look for answers

Computer screen displaying the OpenAI ChatGPT page about optimizing language models for dialogue.
Photo by Jonathan Kemper on Unsplash

ChatGPT doesn’t think… it predicts.

You fire off a question to ChatGPT. You read the reply. Then, perhaps, you type one more thing: “Can you make that sound less like a robot and more like a person?”

I have done it. We all have.

But here is a hard truth about large language models…. When you ask it to sound “more human,” you are simply asking for a better statistical imitation. You are not coaxing a soul out of the machine; you are requesting a more complex mathematical outcome.

Every time you type a question and press Enter, something profoundly fascinating happens behind that blank screen.

ChatGPT does not open a browser. It does not perform a search. It takes your perfectly nuanced human words, converts them into numbers, runs a massive calculation across billions of parameters, and predicts the single most likely reply.

It feels like a true conversation, the kind we have across a coffee table, but it is not. It is math wearing a mask of thought. And once you understand the simple mechanics of that math, you stop guessing how to talk to it, and you start using it like the powerful tool it…

Leave a Reply