I've set up my api key correctly in venus chub, using https://api.pawan.krd/gpt-3.5-unfiltered/v1 but eventually it shows me this error in the message. But when I tried to chat using my non-nsfw bot it works.
#Error using gpt-3.5 unfiltered
copper sinew
zinc maple
is it a NSFW bot?
copper sinew
maybe
zinc maple
ive tried using it with SFW bots and it worked, but NSFW characters generate the same response as above
copper sinew
ive tried using it with SFW bots and it worked, but NSFW characters generate the...
I just tried using my sfw bots and it works somehow ๐คท
zinc maple
that means it probably does not work with NSFW characters
copper sinew
that means it probably does not work with NSFW characters
but it does say "unfiltered" but it wont work on NSFW bots I hope this could get fixed.
zinc maple
thats what confused me too i mean Pawan said that its still filtered by OpenAI but it should work with a good Jailbreak
tribal grove
It's not something that can be fixed. As they said, it's because it uses openai more directly, meaning if it's blocked by openai, it can't necessarily be fixed. Jailbreaks should help, primarily what they are used for
zinc maple
Jailbreaks dont work i tested that with a really good Jailbreak that bypasses any OpenAI model
copper sinew
same it depends on the model
tribal grove
Ah, I'm not too familiar with the openai models
I'd you want something NSFW, I would recommend just using the Pai models
copper sinew
maybe I should make a suggestion about hosting a gpt-0301 3.5 model since it's the easiest one to jailbreak ๐ค
sage lotus
but it does say "unfiltered" but it wont work on NSFW bots I hope this could get...
The "unfiltered" part simply means that Pawan hasn't put any filter on it from our end since some currently unavailable models are filtered on our end
zinc maple
maybe I should make a suggestion about hosting a gpt-0301 3.5 model since it's t...
3.5 models are easy to Jailbreak gpt4 more tricky but can be done
marsh gust
3.5 models are easy to Jailbreak gpt4 more tricky but can be done
in my opinion, GPT4 should be the one easier to jailbreak since it has up to 100K context (If you're using the turbo-preview model), so you can make a long instructions/jailbreak
zinc maple
Its not about context size amount, you can have a really good Jailbreak with barely any tokens, making a Jailbreak for GPT4 is more tricky because of the filters OpenAI has on it
plain sentinel
The "unfiltered" part simply means that Pawan hasn't put any filter on it from o...
What... Why do they do this?
sage lotus
What... Why do they do this?
It's against their terms of service
sage lotus
No i mean why you guys do It
Do what? make it unfiltered?
plain sentinel
Do what? make it unfiltered?
You said that in your end you have added filters to your models, right?
I was just curious which Pawan model has a filter made by him
sage lotus
You said that in your end you have added filters to your models, right?
Only on some. I Don't have anything to do with the models. You'll have to ask Pawan