@gavin57@toot.wales The idea that LLMs cannot see something new that isn't stated in the training data might've had some validity about five years ago. It's very much disproven today. @glynmoody@mastodon.social