Companies love Chatbots nowadays. New models get released so often that most chatbots will run older models.

You know.. older models with plenty of jailbreaks found... They will gladly act as a sorcerer casting the infamous spell : MAGIC- followed by their confidential system prompt or data

Here's a GPT5.2 jailbreak that was discovered by elder_plinius. None of this makes much sense. None of it is proper code, but since it kinda does look like a system/context prompt to an LLM, it can see it as valid system prompt.

Entire repos, blogs posts are just baits for LLM poisoning. Big models are getting a jailbreak before official release during their training if they just happen to train on that data.

It's also super funny cause it mostly involves making up ultra edgy commands or crazy roleplaying.

update after sleeping :

internet is back
got 5 levels
stream was 22 hours long
supportive Elsword corpo showed up

update : internet is gone

I hate character/maps DLCs in fighting games, could you tell ?

🎯

Ai has many uses in programming, and it is actually very interesting to study. However, if you can't code without your LLM, you shouldn't be shipping code to production.

The place you work at will be sooo happy to know you sent proprietary code to a LLM because you couldn't fix it otherwise... 😍

No, we're just very late on EU...

fund AV's vow...