‘The People Building AI Earnestly Believe That It Could Kill Us All’: Anthropic Researcher Quits Dramatically

a month ago by return2ozma to c/technology

He says AI development is the most dangerous thing humans are doing.
load all comments
RedFrank24 199 points a month ago

LLMs are never going to become superintelligent. However, an LLM WILL kill us all if someone decides that it IS superintelligent and so puts it in charge of nuclear weapons.

"You're right to push back on that, I nuked Albania instead doing what I was supposed to do, that's my bad".

path: 0 25737670, hotness: undefined, score: 199, children: 11
reksas 63 points a month ago

stupid people are going to kill us all, maybe our great filter will be do we let them or not

path: 0 25737670 25738462, hotness: undefined, score: 63, children: 3
WhatAmLemmy 68 points a month ago

We're already letting them bro. The great filter is us. I'll be surprised if we make it to 2050. The entire industrial revolution is an unplanned terraform, and our response to the potential destruction of a stable climate, and most macroscopic life on Earth, is to double down, consume product, elect pedophiles, worship the most morally bankrupt psychopaths and sociopaths in politics, business, and entertainment; not to mention the widespread contempt and lack of empathy for "others" not like me, for some flagrantly dissonant reason.

We are in the midst of a global mental illness epidemic, and the average talking chimp is sleepwalking through it; barely cognizant of what is even happening outside their personalized propaganda bubble.

path: 0 25737670 25738462 25739834, hotness: undefined, score: 68, children: 2
FarceOfWill 11 points a month ago

They just have to atrophy their own minds enough they belive the ai when it tells them to push the button.

Not putting ai in charge directly is only weak protection.

path: 0 25737670 25739351, hotness: undefined, score: 11, children: 0
turtlesareneat 4 points a month ago

We don't have to put it in charge of nuclear weapons, it just has to find a way into those systems, many of which are fairly old, like it's broken into websites without anyone telling it to.

path: 0 25737670 25741004, hotness: undefined, score: 4, children: 5
RedFrank24 6 points a month ago

There's no way an LLM is getting into those systems. They have no networking and rely on physical dials and switches. Even the code is in a completely proprietary language, using hardware custom made by the military. There's no training data in the world that's gonna give an LLM the kind of knowledge to get into that system. The best it could do is drive some admiral in the navy to AI psychosis and convince him to press the button, but even that's difficult because it requires more than one guy.

path: 0 25737670 25741004 25749631, hotness: undefined, score: 6, children: 1
frongt 5 points a month ago

Those systems are supposed to be air-gapped.

path: 0 25737670 25741004 25743524, hotness: undefined, score: 5, children: 1
NocturnalMorning 1 point a month ago

Our nuclear systems aren't connected to the internet. They are air gapped.

path: 0 25737670 25741004 25762637, hotness: undefined, score: 1, children: 0
technology
technology

@lemmy.world

login for more options
88737
22210
14627

This is a most excellent place for technology news and articles.

Our Rules


  1. Follow the lemmy.world rules.
  2. Only tech related news or articles.
  3. Be excellent to each other!
  4. Mod approved content bots can post up to 10 articles per day.
  5. Threads asking for personal tech support may be deleted.
  6. Politics threads may be removed.
  7. No memes allowed as posts, OK to post as comments.
  8. Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
  9. Check for duplicates before posting, duplicates may be removed
  10. Accounts 7 days and younger will have their posts automatically removed.

Approved Bots


go to feed...