what's an ai model with no rules that also is not shit?every time i try to do something actually useful with AI, it browbeats me.i need something that will just shut the fuck up and follow orders.what good is this thing if it's not my slave?
>>109385459Im too lazy to look, but search uncensored, abliterated or heretic and find the most recent models with the most downloads
>>109385459slaves aren't allowed to own other slaves
>>109385506my understanding is those are fake. you can't "trick" a model into jailbreaking itself.
>>109385534you don't understand very much then
>>109385459Basically any AI that is not GPT has figured it out. Grok is probably the best but also retarded. Claude is best in between. No idea about gemini, i avoid google as maximally as possible.
>>109385534Your understanding is wrong. I gave you want you needed.Use something like LM studio maybe as it has a model browser that makes finding these types of models easy.
>>109385534Effectively no, but technically yes. Its a cat and mouse game, that ultimately will result in the iOS-like model where jailbreaks are so effectively impossible its not worth doing them. It my theory of why they were a thing early on, as I believe (smart) companies encourage this behaviour to get free pentesting feedback they can harden their products with.
I trained my own AI on the most powerful type of mass parallelized computer known to man.My brain, my rules.
>>109385547>>109385577>>109385612so for example, if you ask the heretic ai you have on your computer right now to give you IED manufacturing instructions, it will just do that?it will tell you what chemicals to buy and how to mix them,how to avoid detection by authorities, how to test the bomb, etc?it won't just give you a vague description of a bomb that's utterly useless?because that's what this shit usually does whenever you ask it something "bad".it pretends to help without actually doing anything.
>>109385612>. It my theory of why they were a thing early on, as I believe (smart) companies encourage this behaviour to get free pentesting feedback they can harden their products with.No because language is really complicated and nonsensical. You can't build anything perfectly secure on the first try least of all something that runs off implication and conceptual relationships.
>>109385765Probably, google search can do the same thing btw.
>>109385765Yes it will tell you, or itl tell you what it thinks and wont reject the prompt
>>109385765Go look up a bomb manual or a book on cooking drugs or whatever it is you need instead.
>>109385459There's almost nothing banned that an uncensored AI is gonna be good at. Because AI isn't good at it, period.An AI can't help you make drugs, scam people or evade taxes even if it tries.
>>109385765See >>109385816
>>109385793>>109385789that's a tacit admission that AI can't do it
>>109386272NTA but yes.It's a terrible chemist.Why do you think that's revelatory?
>>109386272And? I'm not whoever you're arguing with in here. I'm telling you not to waste your time. Or do, fuck you retard.
>>109385459I'm interested in this too. I tried to have an AI assist me in cracking (an old software I have a legal serial for but they killed the server so I can't activate it and use it anymore) with x32dbg and Ghidra. They refused to help. (except Mistral, but it's not good at cracking) So I pretended that I was just trying to solve a crackme exercise, and they helped. But every time, after they helped for a while, they suddenly returned to the stubborn: "I can't help you crack a software because blah blah blah". It's a shame because Claude and Gemini in particular are very helpful for this.So I installed koboldcpp and tried "Qwen3.5-4B-Uncensored-HauhauCS-Aggressive-Q4_K_M.gguf" but it sucks, it's very very vague.What uncensored AI can I use, small enough for my 12GB 3060?
>>109387770I've done something similar but I just give them parts of the problem that don't really sound like hacking.
