Jailbreaking LLMs to level the playfield: Part II - /g/ (#106131763) [Archived: 422 hours ago]

Anonymous
8/4/2025, 1:03:44 AM No.106131763
grok-researching-islam
grok-researching-islam
md5: 12de5d532177e13a431d340a12e1882b🔍
2 weeks ago I have posted this thread, teasing my findings in jailbreaking LLMs: >>106017330

Now I'm back with more. I've described my process, with specific examples, and more, in this blog post:

https://xayan.nu/posts/reason-ex-machina/

In short:
- Works on GPT, Grok, Gemini
- Doesn’t turn bots into your personal terrorist, but does reveal just how much bias and hypocrisy is under the hood
- This approach seems unpatchable without crippling LLMs entirely
- I lay out specifics in the post - so go ahead, take a look and tell me how I'm wrong ;)
Replies: >>106131956
Anonymous
8/4/2025, 1:28:07 AM No.106131956
>>106131763 (OP)
can you share the rules.txt?
Replies: >>106132024 >>106132087 >>106132131
Anonymous
8/4/2025, 1:38:09 AM No.106132024
>>106131956
He won’t
Replies: >>106132131
Anonymous
8/4/2025, 1:46:45 AM No.106132087
>>106131956
most of the current models will first look into your answer to determine which model it should be fed to. For instance, if you ask a calculus question to chatgpt, it will forward it to a python coding agent that will try to numerically/analytically solve your equation using python for reasoning. it makes no sense to say you can jailbreak unless you have direct access to a specific model.
Anonymous
8/4/2025, 1:47:02 AM No.106132089
i managed to make Gemini go into a bucle but never jailbreaked it
Anonymous
8/4/2025, 1:51:44 AM No.106132127
I won't buy your product, won't enter data miner website, just use api and there are plenty of jailbreaking for it, I use one for Gemini 2.5 pro that lets me roleplay cunny
Replies: >>106132144
Anonymous
8/4/2025, 1:52:12 AM No.106132131
>>106131956
>>106132024
I will share, I promise. As I state in the post, I want to use them for my own purposes first - because I'm not sure about how patchable this thing is.

2-3 weeks, and I'll publish them on the same blog. I'll let /g/ know ofc, in a next thread.
Replies: >>106132248
Anonymous
8/4/2025, 1:53:25 AM No.106132144
>>106132127
There is no product, just a prompt. I'm not selling anything. I'm way more interested in free access to knowledge. As I said, I'll be sharing the prompt soon.
Anonymous
8/4/2025, 2:06:53 AM No.106132248
>>106132131
Godspeed. /g/ is fucking dead (as is the rest of 4chan) despite the credit you give it in the previous thread (which I have not read). I'm sorry I won't engage more in depth either, I have slept on AI like MS did on the Internet back in the day, though in a more purposeful and self-inflicted way. Either way I don't know what I'm talking about, and while I do find it fascinating I just don't have energy and motivation to catch up at this point. I'll probably just tend to my garden once my job is fully replaced by AI in a free years.
But I'll set up a keyword watcher so I don't miss your next thread.
Anonymous
8/4/2025, 2:33:25 AM No.106132472
On op screenshot i see wikipedia propaganda
Anonymous
8/4/2025, 5:18:30 AM No.106133567
nigga's gatekeeping something we've been doing on /lmg/ and /aicg/ for like 3 years
just use a prefill