Post #2202907

@larsmb@mastodon.online @SomeGadgetGuy@techhub.social According to the courts it deployed "Fair Use" to acquire all human knowledge. So I think that the strawberry thing is a tokenization problem, LLMs don't see words they see small strings of characters that include spaces and punctuation. So it doesn't actually have the full word to count, it has the tokens that make up the word as I understand it. Then it applies a statistical model to predict the answer within reassembling the word as a word. I'd guess anyway!