My experience with the New text model (feedback)
kris_krovish @ kris_krovish @lemmy.world Posts 0Comments 9Joined 3 wk. ago
This. Completely agree. That hyper-fixation is just out of this world - fine, it was present also in the older model, when the characters were fixated on secondary plot, but this time... oh man. And I'm feeling like I'm not in control anymore. I write stories, I simply know where the story escalates, but this new model is like 'I'm gonna hallucinate something and you are gonna play my game instead.' I switched to Silly Tavern instead, which has soooo many possibilities, so many models to choose from, that I'm not even mad, if I ever choose to pay for them. Still exploring that platform.
Something is either even more broken of fixing at the background. Some commands started to appear like this one: 'BREAK_BAD_PATTERNS: Avoid purple prose. Lean into character authenticity.' and then the character response follows.
Created a fresh character. Carefully crafted, great choice of words. But when they started speaking about I quote: 'bombs strapped to orphans', without before mentioning ANYTHING even remotely close to that, it shell-shocked me. However, it started to produce actual longer and semi-meaningful sentences - so hooray I guess? - but no. Still completely rude, bringing a knife everywhere and threatening me to 'make a keychain out of my spine'? Are you serious?! What is this AI trained on?! Freakin' Library Of Bloodiest Books Ever? I pass. I'll wait until 30th of September. And if the AI is not fixed by then, I'll erase every memory from my brain of ever knowing perchance as a platform. Or maybe it's a sick artistic joke? I'm not amused by that, seriously. I know, it's free, so I should not be expecting miracles, but come on! AI is already trying to murder us, guys! :) I don't expect it to be fixed soon, but a word from dev might be helpful. Am I asking too much? If the new model stays - I'm out immediately. If it will be fixed in the future - fine, I'll wait until it will be usable. Maybe. But seriously, I should have spent those 2-3 hours not testing this crappy model, but doing something meaningful, really. And many hours before that. Have a good day.
yep, my conversation also turns to blood bath. At this point I'm out! Wake me up when the AI's fixed. I just can't watch that murder to my characters anymore.
My experience is that the reality is almost non-existent now. My character is suddenly bathing in water - in a pizzeria - not explained where the water comes from - speaking about bombs and sedatives, suddenly forgot that Norwegian accent, which they did not had in the first place, so I -as a user- called another character, which is supposed to be on the other side of the town - immediately they come to the pizzeria right away, calling me an idiot for no apparent reason, sentences are 1-4 words long, and so on. Maybe I should use that for a slipstream-like parody of a story! :)
Edit: narrator just gives up - he's copy-pasting my exact words. Edit 2: I give up. That conversation turns to blood bath, I can't work on that anymore. Deleted it.
It is hilarious, holy... :) In my case I have to stop generating more text to my stories, since I've not found the tweaks around this new model yet and my characters are ruined for now. I mean, I don't mind Naked Gun, but that horror and gross stuff - and in my case swearing from my cutest characters - there should be a way around that, right? :) I hope, we get an update - or some words from dev - until that - we wait. I have another project opened up, where I don't need perchance's AI help since it's in my native language.
Sure. Yes, with all respect to dev we shall be patient and give him more intel about this new AI model, which even makes stuff up. One of my character suddenly have a Norwegian accent, even though, there's simply no such information in description about them. They have blond hair and blue eyes - so must be Norwegian! :) . And sure - they turn rude as well, never before they have told the word "idiot" to other character, they suppose to adore. So, yep - all we have to do, is just wait for an update.
Agree! It's not fun to interact with such rude AI, which made the AI chat unusable and less enjoyable.
Thanks. ST looks a bit chaotic at first, but it's still just a front-end for the AI model in the background. Well - with ST you kinda feel like a mad scientist - you are much in control and see what the characters are talking about if you play with hundreds of checkboxes and descriptions. Narrator is kinda missing, but you can have like an Author's Note, so the plot can thicken further, but without descriptions about the scene. However, one of Microsoft's model, which kinda resembles perchance's old model - also ruined my character! It started to escalate and escalate- not by violence this time, but with the respect to me as a user- she became like a girl-boss at the end. And today I might have figured it out what went wrong - Depth- which is similar to Long memory. The character responses were generated from only 4 previous messages. That's why every time I say something playful or teasing, she wanted to overcompensate that and the snowball effect wrecked the whole conversation :). Also - I can't see the possibility to prompt the AI's response. Sure, you can edit it after the answer is generated, but giving the character a bit of a nudge before the answer is generated- maybe it's possible, but I did not find it yet. And by the way - I can confirm than DeepSeek is violent as heck :)