Post #3214551

@Nirik@hooves.social Hi! I decided to reply because your post is right up my alley. You're correct that translators work like primitive neural networks. The main difference is that LLMs were trained on much more data. This gave them emergence. Thanks to the massive connections between concepts, these models build their own world view and can "understand" the core meaning of a text. Traditional translators were only trained on pairs of words and phrases, which is why they make mistakes more often.