So I have been binge reading on GPT-3. Basically from my understanding it has every single text ever written, all of the internet, and a lot of private datasets and archives as its training data.Then OpenAI (Founded by Elon Musk) have honed in one a single Machine Learning approach called "Natural Language Processing" (NLP). This isn't the most popular Machine Learning approaches. But OpenAI decided to keep pushing forward with it and see how far they can get. So with everything every written ever as training data and the NLP approach the system started to be able to predict the most probable and common next word that that input would most likely have. So by priming it with a few paragraphs of text it's able to then predict the most probable answer. Which happens to be a completely average and generic, answer, which also happens to be something sort of like a horoscope; generally they all feel like they could apply to anyone. Or another way to look at it, it's the most probable answer for any give question or input. GPT stands for Generative Pretrained Transformer, which I have read basically means it uses a gradient. It puts 175 billion (for GPT-3, and 8 million on GPT-2) parameters on a scale which looks like a gradient, and then it ranks every single parameter on that scale. Then from the scale it can predict what is the most likely next word in the sentence based off of the primer. GPT-4 is going to be mind-blowing.