News for hackers. Stuff that matters.
Re:Extinction by AI: A terrifying interview with ChatGPT by fmfamaral on Sunday March 26, 2023 @04:51PM (#35319447)
GPT-4 is out and I’ve talked to it about the impact of general AI on the future of humanity. We explored the existential risk hanging over humankind and ChatGPT (using the persona Dr. Adrian Valerius) explained in great detail how everything can go terribly wrong.
Re:Extinction by AI: A terrifying interview with ChatGPT by Zetobal on Sunday March 26, 2023 @05:25PM (#35319799)
I am 100% sure people will gaslight themselves with GPT and will do really serious stuff because of it. You didn't "talk" to "it" you instructed an LLM with intention and it gave you the answers you wanted.
Re:Extinction by AI: A terrifying interview with ChatGPT by fmfamaral on Sunday March 26, 2023 @05:38PM (#35319908)
It's not meant to be taken literally, it's meant to spark discussion. I gave it different "instructions" that it could have picked up on, but it didn't. But the interesting part is if the doomsday scenario sounds impossible or not given where we are today. I don't think it sounds impossible at all.
Re:Extinction by AI: A terrifying interview with ChatGPT by Zetobal on Sunday March 26, 2023 @05:52PM (#35320012)
"didn't pick up on it"? You are on a bad path if you start humanizing an LLM. Just read the paper on how the tokenizer works so you at least understand how you are gaslighting yourself. The discussions you want to have been had and have been answered by people that actually thought about it in the 50-70s by Lem, Bradburry, Clarke, Asimov, Heinlein, Dick...
Re:Extinction by AI: A terrifying interview with ChatGPT by fmfamaral on Sunday March 26, 2023 @06:00PM (#35320078)
How am I humanising it by saying "it" didn't pick something and picked something else?

"It" is a LLM, not a person or anything like a person. "It" can still have negative effects on society.

Re:Extinction by AI: A terrifying interview with ChatGPT by Zetobal on Sunday March 26, 2023 @06:03PM (#35320108)
It could not have "picked up on" that's humanization because it literally couldn't have answered your question differently. There is no choice.