> Effectively, they (LLMs) can't think new thoughts. This is true only if you assume that combining existing thought patterns is not new thinking. If they can'

> Effectively, they (LLMs) can't think new thoughts. This is true only if you assume that combining existing thought patterns is not new thinking. If they can't learn a certain pattern from training data, indeed they would be stuck. However, their training data keeps growing and updating, allowing each updated version to learn more patterns.