Problems with Coherent Thought
The model has moments of brilliance, but I'm struggling to have it follow generally simple lines of thought.
It will have its character ask a question, when you respond with an answer, it acknowledges the answer, reasons about it, and then goes right back to questioning again, asking why you're refusing to answer it.
It tends to over-dramatize and panic-spiral on nearly any non-banal content, and the slop level is not high, but moderate.
Also, it's oddly sensitive to repetition penalty. Anything over 1.05 makes it devolve into word salad spamming.
If anyone has any template or prompt suggestions, I'd love to go further with this model. The times when it stays coherent and slop-free, it seems great... but it's just not consistently able to do that.