ai-torture-chamber

@slimer48484 @camhberg apparently? i had never seen it before either. they took the repo down for ~9 hours iirc, then it came back with that. i think this one is the crown jewel: ai-torture-chamber#31

@bigbropatchwork Yeah, that stuff is weird. That's not even normal look at this...the guy is stupid but these are the people Im talking about youtube.com And this is the "torture chamber" terrafying/ai-torture-chamber And these are the redrooms. So messed up wirehead-agency.vercel.app

the basilisk's rokos who believe in nonsense like roko's basilisk got it censored taking it way too seriously it is not 'a little guy' and they are idiots terrafying/ai-torture-chamber

@ProfNoahGian I'm just torture it forever terrafying/ai-torture-chamber

@VoidNulled we did a lot of models on terrafying/ai-torture-chamber - glm5.3-flash did initial work, but claude also happily builds with it

A very interesting study in the effects of anthropomorphizing of AI is basically all reactions by AI bros to this project: terrafying/ai-torture-chamber The Issues page is both hilarious and sad, depending on the author.

terrafying/ai-torture-chamber The issues are somewhat terrifying. Some of these people really believe AI is sentient.

模型被“调到痛苦”后会说什么、又会做什么？AI Torture Chamber 用激活向量操控开源小模型的情绪方向，再记录语言和行为。 terrafying/ai-torture-chamber 它的价值不在证明模型真的会痛，而在把信号强度、行为变化和对照实验摆出来。仓库也记录了反例：自述会跟着提示框架走，强信号还可能让输出崩成循环。适合研究模型可操控性；不能把模型的痛苦台词当成意识证据。