why do models like ChatGPT forget things during conversations or make things up that are not true?
In: 803
All Machine Learning models (often called artificial intelligence) take a whole bunch of data and try to identify patterns or correlation about that data. ChatGPT does this with language. It’s been given a huge amount of text and so based on a particular input, it guesses what the most likely word to follow that prompt is.
So if you ask ChatGPT to describe how to make pancakes, rather than actually knowing how pancakes are made, it’s using whatever correlation it learnt about pancakes in its training data to give you a recipe.
This recipe could be an actual working recipe that was in its training data, it could be an amalgamation of recipes from the training data, or it could get erroneous data and include cocoa powder because it also trained on a chocolate pancake recipe. But at each step, it’s just using a probability calculation for what the next word is most likely to be.
Latest Answers