I love it, but theres just a few nitpicks...
First off let me praise that this model is beautiful. It makes characters feel very real and its great at picking up and rolling with whatever random event you throw at it. It’s so good at pointing out the little details that make the RP feel like it doesn't just exist in some kind of void. It’s also surprisingly great at word play and it can come up with some pretty hilarious insults when portraying a rude or snarky character.
I’m coming from Starcannon and this is undoubtedly better, However there are a few very subtle things that make this really hard to consider as a new daily driver:
Chronos gold was absolutely vicious with its ability to describe gore and horror in general, but that trait seems to be lost in MagMell. I was really looking forward to that...
There is a very subtle but annoying amount of GPT-ism. By that I mean leaving a few purple closing sentence to every paragraph that I’m usually forced to cut out or it'll only get worse and lace purple prose throughout future generations, as well as a tendency to be TOO on the nose when it comes to things that should be ‘show don't tell’ such as a tsundere character mentioning that they are a tsundere, or a motherly character purple prosing their motherly nature.
Short responses. I love really long and verbose responses at around 500 and it can for the first few, but then gradually becomes shorter until I can never seen to consistently get past 250 especially late game.
And one I've noticed was the lack of randomness when regenerating. I have no idea if it’s just my settings, but whenever I degenerate, it’s as if it’s using the same seed. Everything is different but just the same enough to not be a new branch in the story. It takes at least 4 re-rolls to finally see something that doesn't contain one of the prior dialogues or thoughts.
If I could request you add more weight to gold?
Updates to this model with the way it's structured are annoying to do, but I do want to take another run at this since my org Alfitaria has been brainstorming some ways to help with its long-context performance as well. When that turns into something, I'll also try integrating your suggestion!
A lot of this tracks onto Nemo-isms as well; what settings are you using? I might be able to make some suggestions. (I personally just use the Universal-Light preset on SillyTavern.)
You see,this model does not use google NOR internet. Which leads to only one solution,you have to make the character card,it listens very well you just have to do it yourself. Most and best way is by copying the whole wiki of the character and the things he knows about. As example if it's Yamato from one piece then you should copy her wiki and the wiki of Kozuki Oden,yes it's hard because you also need to do prompts so it can understand what she must know and what she MUSN'T.
You see,this model does not use google NOR internet. Which leads to only one solution,you have to make the character card,it listens very well you just have to do it yourself. Most and best way is by copying the whole wiki of the character and the things he knows about. As example if it's Yamato from one piece then you should copy her wiki and the wiki of Kozuki Oden,yes it's hard because you also need to do prompts so it can understand what she must know and what she MUSN'T.
That is horrible character card practice. you should never just copy-paste a wiki. Its always best to write the info by hand, and by hand i do not mean let another AI write it for you. Dont be lazy.
Im here in the future where this model has become my daily driver and ive used it enough to understand its particular quirks and limitations. It has a very crippling positivity bias but more often than not this only has an effect on dark scenarios, this opinion and pet peeve still hasn't changed.
One thing i will mention is that it isnt very sensitive to the system prompt and by extension character cards, which could be good or bad depending on your preference or project. unlike some models, the personality, history, and rules feel more like subtle influences than direct instructions. if i want a very narrow aspect of a character or plotline, i more often than not need to directly OOC or inject at a depth of at least 1 or it wont have much urgency. for example, saying that a character has a history of drug use or that they have a stronger hatred for one item over their other listed hates. While it can remember these details wonderfully, it wont give them too much weight unless you specifically place them in high priority (specifically in character notes at depth 4 works great). same with scenario, it feels more like a general grounding outline rather than "this is where the story is going this is what we're doing right now".
Ive spent the past year tweaking my system prompt and settings and have settled on a setup that does exactly what i want with the perfect amount of prose. every issue ive had in my first message (aside from the lack of chronos-gold brutality but ive come to terms) has been either fixed or ive just sort of gotten used to em. I use this model daily and its genuinely been hard finding a replacement. Ive ended up developing an elaborate system that instead implies we are writing a book instead of playing roleplay to change how the prose looks and works in my favor.
You see,this model does not use google NOR internet. Which leads to only one solution,you have to make the character card,it listens very well you just have to do it yourself. Most and best way is by copying the whole wiki of the character and the things he knows about. As example if it's Yamato from one piece then you should copy her wiki and the wiki of Kozuki Oden,yes it's hard because you also need to do prompts so it can understand what she must know and what she MUSN'T.
Congratulations, you've sussed out that this is a model from 2024 before anybody did that. ChatGPT barely did that at the time. You are in fact going to have to craft your own prompts and cards
You see,this model does not use google NOR internet. Which leads to only one solution,you have to make the character card,it listens very well you just have to do it yourself. Most and best way is by copying the whole wiki of the character and the things he knows about. As example if it's Yamato from one piece then you should copy her wiki and the wiki of Kozuki Oden,yes it's hard because you also need to do prompts so it can understand what she must know and what she MUSN'T.
Congratulations, you've sussed out that this is a model from 2024 before anybody did that. ChatGPT barely did that at the time. You are in fact going to have to craft your own prompts and cards
im a cave man from said year, what does google and the internet have to do with this? i rarely use API or ChatGPT itself.
edit never mind i understand it. using the internet for the info of a character card. thats actually crazy i never thought about that. doesnt sound very practical to me however as none of that info is stored in any local capacity i assume.
hey so it turns out I'm actually a huge idiot who thought this was a new issue for some reason because I had just opened my eyes when i'd read dfdadsadsa's comment and then proceeded to completely misinterpret it LKSDJFHGKLJ
feel free to ignore me r/n
At first I was thinking about not replying to this but I just can't help myself and had to say the following:
This is literally, to date, still my favorite model ever that I have come across! (Oh and @inflatebot I literally almost have finished the vision tower addition to this model ;) . I was thinking vision tower + light updated training for catching it up a couple years = perfection.
Though correct me if you have other opinions, I'm open to whatever critisisms or banter regarding this.
@IZA09 , I would agree with the other fellow that stated the model indeed is a very capable listener if you provide it a baseline to go off of. Literally I can confirm that if you imbue a "persona" onto it - it never stops absolutely blowing my expectations out of the water, very little repeating (if ever) and very capable with linguistic skills. Though I do realize not everyone enjoys prompt "engineering" and it can most certainly be tricky sometimes.
@dfdadsadsa , striaght up upload a whole wiki to drive the direction? That's impressive I had never tried something so extensive. Typically I'd create maybe 3-5 paragraphs and of course carefuly revise it until I was seeing the desired outcome. Might I ask, how are your experiences with dropping such a massive amount of text as that of a wiki vs creating shorter "character sheets" and such?
My apologies in advance if this message annoys anyone. Just had to give a shout out to what I still think is one of the best models out there going on two years now.