I got bored and wrote about AI risk 😂😂😂
Conversation
I think said as much in our episode.
He doesn't think an LLM will destroy humanity. He thinks an AGI will.
Do you have an argument against this?
5
48
No, it's just impossible to know, which is why people drive themselves up the wall thinking about it
3
1
17
Good to know that anything not directly observable is unknowable until after it happens! Have you tried applying this amazing insight evenhandedly to other fields where outsiders are skeptical that anybody can really know anything, like economics?
4
2
111
Also, by "impossible to know" do you mean that the only reasonable probabilities to assign are 50-50?
1
32
No. When you're dealing with Knightian uncertainty, there's no universally accepted mechanism for determining what reasonable probabilities are.
2
10
So, like: "Impossible to know! No authority can put any numbers on it! But the numbers are small and nobody should worry."
3
48
Well, if someone does say the numbers are small and nobody should worry, there's really no data to contradict (or support) them.
5
1
5
Soooo... the state of debate is then...
Eliezer: "I think it is guessable that we're all going to die in the future, for reasons and arguments X, Y, Z..."
Noah: "Don't worry, we're not all going to die today! As for the future, impossible to know."
Noah, the expected value of an event with magnitude of -∞ and infinitesimal non-zero probability is -∞.
1
1
Irrelevant, since nothing at hand is infinite; doubly irrelevant, since I'm claiming the reasonable probability is over 50%.
1
11
Show replies
3
15
Show replies
"We're all going to die at the hands of AI" is the traditional opinion of AI.
The evidence over the last year is AI becoming extremely powerful, whist also being wholly benign and wholly alignable suggests that our limbic system isn't the best judge of these things.
Isn’t #AI misleading b/c:
It’s not intelligent;
It can’t reason or be reasoned with;
It can’t learn despite labels as “machine learning.”
At core, AI compiles data & runs permutations for best outcomes, and stores & recalls prior outcomes.
AI ever halt an execution & correct?
GIF
read image description
ALT
I like Noah's stuff in general, but on AI and climate risk he's hopelessly optimistic to the point of them being major blindspots. I trust the late Stephen Hawking more
1
Show more replies







