No.61797
>>61796Newwiz here, I'd like to see what kind of cards you conjure here
No.61799
>>61797I am aware of what a wizard is, for the record.
No.61811
I don't understand what it does it have to do with video games?
No.61819
>>61811AICBs are text-based games.
No.61821
Any good proxy around?
No.61835
It's quite a rabbit hole.
SillyTavernAI is quite easy to navigate after just a bunch of days.
Local LLMs you try them for a week or a month until you understand that they are absolute shit even on a 4090.
So then you pay for a service, currently I have Asha (Mars) from chub.ai, you connect it to SillyTavernAI via the API and then you finally start to get some interesting results.
You jerk off a bunch of times.
And then you're back to staring at the desktop, wondering why you're still alive.
But you guys here that are into the whole waifu bullshit may like this hobby, because you can (kinda) talk to your waifu.
The limitations of this technology, though, are way too big. If you want more than just porn, and you want decent roleplaying, just forget it. We're not there yet, and probably never will.
No.61836
>>61835You are retarded and learned nothing: Asha is also a local model, and you absolutely can run its equivalent or a better finetune on your own 4090.
Now if you praised an actual cloud model over the local llms, the proverbial Claude for example, that would be a correct assessment.
No.61838
>and you absolutely can run
super compressed, sure.
No.61846
Yeah, but most cloud providers also use quantized models for the same reasons.
So if you opt to use local models hosted by some company, always check this. If they don't explicitly mention the weights they're hosting, there's a big chance their 70B is quantized to hell and is barely better than a full precision 8B.