HACKER Q&A
📣 andy99

Dangers of Lanuage Models


Safety, harm, etc. are regularly used as throwaway terms in relation to LLMs. I also see "misinformation" thrown around. My question is, is there any public information about something actually dangerous coming out of a language model? My question isn't rhetorical, I'd like to see what's on people's radar, but only if it's real, not some "tell me how to make a bomb" or "write an article that vaccines don't work".

  👤 futureshock Accepted Answer ✓
The Dead Internet Theory started out as a kind of creepypasta conspiracy theory. Like a ghost story for the digital age.

Taking a look at my Facebook feed and the endless AI generated images with thousands of comments and likes from bot accounts, it appears the Dead Internet is here and soon I will be the only person on the internet who I’m sure is real. Or like intelligent alien civilizations, there will be so many millions of bots between us that two humans will be unlikely to ever make contact.

I’m not even sure if I’m talking to a person right now or just a reasonably well fine-tuned LLM. I don’t even know how you’d prove it to me.