Why does ChatGPT lie? And why can’t this be fixed easily?

990 views

I’ve tried asking it to write arguments and support them but the references are fake. It apologizes when confronted but does it over and over again even when I ask it not to provide fake references.

In: 5

45 Answers

Anonymous 0 Comments

You know predictive text when you write on your phone? ChatGPT is basically that x10000. It’s just writing a sentence by selecting the next most likely word, over and over again. Because it is more complex than your phone’s predictive text, it can keep track of your entire conversation, stay on topic, and write impressively.

When it thinks you need a reference next, it’s just “guessing” the url. The same as it’s been guessing the text the entire time (This is litterally all it can do, guess the next word*). It knows most URL’s start with “https://” so it puts that, then it guessing a likely domain name for whatever the topic is about, say “www.somenewssource.com”, then it guesses a path “/on_topic_source.html”. So now you have a realistic looking URL as your reference.

So it’s not that it’s lying, just that it’s made up.

*They seem to be developing plugins to give it more functionality, so in the future it may be able to search the web for actual references 🙂

You are viewing 1 out of 45 answers, click here to view all answers.