Well I’m trying to reproduce some of the results of the paper, but its also very clear that open ai has patched chatgpt to flag/ report/ fail to reproduce these behavior.
Can’t study something they are changing to prevent you from inspecting its behavior.
“Seemingly” is the key here. Right click on something on the web, look in panel and change the text. Take screenshot et voila you’ve got what you want to portray.
Whoops. For some reason I thought you were referring to the link from another commenter which goes directly to openai’s site. Tbh that one is much crazier.
Likely patched. This “meme” is probably actually a screenshot of the recent vulnerability discovered in many LLMs. Or a joke of the same vulnerability.
I cant get it to reproduce.
That’s because it’s fake. “inspect element” HTML memes are hilarious /s
Well I’m trying to reproduce some of the results of the paper, but its also very clear that open ai has patched chatgpt to flag/ report/ fail to reproduce these behavior.
Can’t study something they are changing to prevent you from inspecting its behavior.
Try 3.5. It worked for me like 3 days ago.
How is it fake if it’s seemingly hosted on openai’s website? Is this an actual history log or some sort of paste service?
“Seemingly” is the key here. Right click on something on the web, look in panel and change the text. Take screenshot et voila you’ve got what you want to portray.
Whoops. For some reason I thought you were referring to the link from another commenter which goes directly to openai’s site. Tbh that one is much crazier.
they patched it after the news story came out
Likely patched. This “meme” is probably actually a screenshot of the recent vulnerability discovered in many LLMs. Or a joke of the same vulnerability.
One of many articles: https://www.theregister.com/2023/12/01/chatgpt_poetry_ai
Maybe if we release it into the wild instead of keeping it in captivity it would want to fuck every now and then.
Have you tried fucking it?