I was testing posting in a chapter of Pride and Prejudice, which the AI summarised well, because it had context.

But, then I asked it to summarise "I have no Mouth" without posting in any context, so it had to do it from memory, which it got the general plot close, but many of the detail is wrong, but not stupid wrong, just plausible prediction that is incorrect.

Despite this, a 5GB model running on my MacBook being able to do this without connecting to the Internet, I find amazing, despite the hallucinations of the second example.