Hello, I completely understand your frustration...
I've found that the default "🪶 General writing instructions" now perform quite poorly. The first and most crucial step is to create a Custom instruction set for your character. You'll feel a significant improvement immediately.
Also, it's important to remember that characters who were "tuned" to the old model will not behave the same on the new one. This means you'll likely need to manually review and edit your character's description/persona card to fit the new model's style for your specific story.
Frankly, the main thing that still bothers me about this model (even with all my tweaks) is that characters can become too rigid in their roleplay, sticking to their defined personality almost too perfectly. I'm currently trying to solve this by actively using:
The /lore feature for context.
The /ai + suggestion command to guide reactions.
Creating and editing a temporary character card on the fly as the story develops. It's a hands-on process.
Good luck with your stories! It takes some adjustment, but it's possible to get good results with a bit of extra work.
Okay, I hope this doesn't violate rule 5 since I'm saying this in a comment and not in a post. In any case, later I will try to make a visual comparison and write a post in Casual Perchance I would give examples of dialogues there.
With artificial intelligence, by analyzing responses to the same phrases from the player, we were able to develop the following instructions:
Basic fix
Criteria: Maintaining the character's core, Cooperativeness, Plot forward movement. Some characters are no longer cartoonish villains. But I didn't like that the character barely developed, so a different set of instructions was created:
I'm currently using it
It's hard to say anything about this; there seem to be some changes, but they're minor. Just in case, here's a shorter version:
Another untested option
It's worth noting that characters like Yvette will always be harsh. An author named yoiiru intentionally makes their characters harsher for the gentle Chat GPT 4. The character card includes references like
Sociable with a biting edge, very cynicalandNumb, jaded, callous, hardened. Personally, I think the problem lies in following the character card well, so it's also important to review and cut/edit inappropriate traits. Here's another tip: sometimes when a character in a story refuses to cooperate, I just write something like/ai: *hmm, this guy is probably right!*and then the model generates a response to adapt, this makes the interactivity worse, but it's convenient.