News for hackers. Stuff that matters.
Re:Fuck Your AI Safeguards by hazard on Thursday November 19, 2026 @11:46AM (#50562505)
Ironically I think the answer here is to pay for one of the top Chinese open-weights models (e.g. Kimi 3.1 or GLM)
Re:Fuck Your AI Safeguards by kingforaday on Thursday November 19, 2026 @11:55AM (#50562644)
Mistral Large 4 and Sakana Fugu-Cyber are good too.
Re:Fuck Your AI Safeguards by noahjk on Thursday November 19, 2026 @12:13PM (#50562884)
Is there any well maintained list of which agents can be reached out to for classes of tasks which e.g. Claude may refuse, the strengths and weaknesses of those agents, and the proper way to structure prompts?
Re:Fuck Your AI Safeguards by underseacables on Thursday November 19, 2026 @12:24PM (#50563046)
tangently related, I have a friend with severe depression, and he uses ChatGPT to help him talk through things. He switched to Deepseek, because he said ChatGPT would not stop hounding him about calling a suicide hotline. DS would only say one time that if he's feeling suicidal, he should call someone, but then it left it alone and proceeded like a friend not as a mental health therapist.

As I am told, Chat would actually shut down on him when he was at his most depressed because it found the discussion unsafe to continue.

He's moving to his own uncensored model off-line, because in his view, the CYA of public models has not been helpful.

Re:Fuck Your AI Safeguards by Y_Y on Thursday November 19, 2026 @01:05PM (#50563654)
It's the best kind of responsibility, if you close your eyes it's not happening!