Sometimes when prompted to generate text, (for example letter) it goes into an infinite cycle of generating a one line template response like:
“Here’s a straightforward and tailored letter emphasizing your experience with …”
“Here’s your personalized letter focused on your …”
“Here’s a concise, tailored letter specifically emphasizing your expertise with …”
It is very interesting to me, how it behaves like a small language model that is not properly finetuned. But what might be the reason for this? Any thoughts?
and it has occurred for me following extensive chat, when cgpt shared with me the notion of “instability” after a large amount of data is in the single convo