It's a language model. It's not "taught" anything. Stop anthropmorphizing the product.
They scrape the internet. They filter it down to the sites they "want." They pay people in Africa to "tag" the content with context. The language model does a highly dimensional search through this data set.
That's it. It doesn't "know" anything. It just has data that has been created for it. Even then, it's not good enough, so they use a weaker language model to filter the output of the stronger language model, so it only says what they expect it to say.
You're so far away from the "machine knowing something" it's not even funny. It's just the Wizard of Oz messing with you. On purpose.
It's a language model. It's not "taught" anything. Stop anthropmorphizing the product.
They scrape the internet. They filter it down to the sites they "want." They pay people in Africa to "tag" the content with context. The language model does a highly dimensional search through this data set.
That's it. It doesn't "know" anything. It just has data that has been created for it. Even then, it's not good enough, so they use a weaker language model to filter the output of the stronger language model, so it only says what they expect it to say.
You're so far away from the "machine knowing something" it's not even funny. It's just the Wizard of Oz messing with you. On purpose.