It looks like the model got changed again; will it be possible to revert to the older AI model or will the current one be the new default? I personally prefer the older model as this newer one seems to despise dialogue and, when it does use dialogue, tends to sound like someone trying their hardest to be mystical and spout nonsense. The current AI seems overly verbose. It overcomplicates descriptions and just generally uses a lot of words to say something that could've been done in one sentence or less, often seems to brush off the player's given prompts in favor of elaborate descriptions about the ambiance of the area you're in, and it seems overall just less refined for this type of application than the older model. I've straight up prompted to greet an entity and the AI will return something akin to "You offer a brief acknowledgement and, if they notice, they offer a nod in return. The air is professional and quiet and you continue on your way." and railroad you in a direction you don't necessarily wish to go while trying to avoid any dialogue at all.
Renting GPUs to do anything is super expensive right now, especially so if the task is AI related. The model change was to try and get something that won't be unreasonably expensive to pay for. Best option if the current quality isn't good enough is to either create an account on an LLM hosting site or try out local stuff. Small LLMs able to run on <4GB of VRAM are surprisingly good for the hardware/effort required. The desktop app runs models natively and has a few suggestions of small models to try out which can be downloaded directly in the app itself, no technical skill required.
I’d say I’m also struggling with the new model. It doesn’t like dialogue and it seems insistent to be overly verbose with big words without actually following directions. I instruct my character to examine the clothes they’re wearing and the AI has them arbitrarily poke holes in the fabric and take it off.