Why can't language models come up with new ideas?

&gt; Humans are thus also bounded by their training data. I cannot have an idea that’s totally separate from anything I’ve ever read or experienced.<p>Is that just an assumption or backed by science?