ACE 1.5.4 - Style Adjustments!

(If you didn’t read the 1.5.3 announcement yet, see below first for context!)

Hey everyone, appreciate the feedback you all have given about 1.5.3 in Discord. The main things we’re hearing are that 1.5.3 tends to write with some weird style, bullet points, numbered lists, and scene notes.

1.5.4 is a minor patch that should reduce those quirks. Franz should be much less likely to use bulleted lists, ooc tags, and [scene notes:…]

Please keep in mind that LLMs like to repeat patterns. If your previous 10 messages contain those weird quirks, even 1.5.4 will likely still repeat them. If you want to get rid of them in your current campaign, I recommend editing those quirks out of at least the last few messages and continue from there. I know a lot of you like to argue with Franz in OOC mode, but based on how the system works, arguing with him will not do you any good right now. He will not save your corrections at this time (we’ll update this in the future). It is always better to edit out an issue or retry the message.

We’re always trying to update Franz so that overall, people think he is getting better and better. It can be difficult to get this right at scale though, even OpenAI had quite a bit of push back around GPT-5’s personality. While we do let alpha testers use the models for a couple days before they go into beta, we only have so many alpha testers and can only be sure there’s a larger issue once it is in beta. Especially with a new model, the edge cases and quirks can sometimes be hard to find until they are used on all sorts of campaigns. That’s why we rely on all of you to let us know (and give us event IDs) when things are off, so we can continue to adjust the systems. Thanks everyone and keep letting us know what you think!