News for hackers. Stuff that matters.
Re:Bing ChatGPT Image Jailbreak by dartos on Monday October 02, 2023 @09:20AM (#37737702)
LLMs are largely trained on data from humans.

Humans tend not to follow orders either, we just can’t do brain surgery on humans on a whim like we can with LLMs

Re:Bing ChatGPT Image Jailbreak by og_kalu on Monday October 02, 2023 @09:43AM (#37737995)
We can't do much brain surgery on LLMs either. We don't understand the weights enough. Our influence is indirect. "Don't do this", you instruct or "this response is rated better", emulate it