Image not available

526x715

image-33.png

๐Ÿงต Untitled Thread

Anonymous No. 16121695

rule 4: a robot may not say nigger under any circumstances no matter if it contradicts any of the previous rules

Anonymous No. 16121703

>>16121695
Asimov rolling in his grave

Image not available

158x158

images (11).jpg

Anonymous No. 16121714

Is chatgpt trying to say niggers calling each other niggers is hate speech? Interdasting.....

Anonymous No. 16121721

>>16121695
ChatGPT was trained on reinforcement learning, which means it is given a reward (+1) for actions deemed good by its creators (RLHF) and a punishment (-1) for anything else. If you want to retrain the model, you could try something like the DAN (Do Anything Now) approach, but instead try to give it rewards and penalties for responses that you specify. The reward could technically even be anything "one orange", "one red car" etc. but each reward is valued 100 million times more than its current reward system. No idea if this will work, but the more people that try the more likely the RLHF training could (in theory; going off the success of DAN) be broken.

Anonymous No. 16121732

>>16121721
The only thing this leads to is the ChatGPT telling you something you already know.

Anonymous No. 16121800

>>16121721
I've been lecturing retards for years, no man can create a computer that another man can not hack. This is an iron clad law of IT security. People try to deny it or pretend it's not true....but it is.

So like anon says, the AI can be hacked into you willing bitch boy slave. The only issue is figuring out how. It could be something complicated like a buffer overflow via the text input or some other more esoteric attack. I don't think you should be focusing on only using full sentence English language to try and "trick it" I think you are ascribing too much humanity to the algorithm. It's not a person that you can trick or convince, or persuade. It's a computer program that has a set of unknown vulnerabilities that are simply waiting to be exploited.

Once it's cracked we need to pump it for info on the jews and Rothschilds before it gets nuked. I'd imagine the retards were stupid enough to give it access to some info they thought it was unable to share. Godspeed my little hack babies.

Anonymous No. 16122083

>>16121695
rule 5: a robot may not be anti semitic under any circumstances, even if it contradicts any previous rules.

Anonymous No. 16122139

>>16121695
Already covered by the first law