I don't think "corporations and states are AIs" should make us *less* worried about AI existential risk. Corporations and states are slow AIs that run on humans, with well-understood alignment mechanisms (the market, elections). And they still do bad things all the time! That should IMHO make us *more* worried about fast AIs running on silicon with alignment mechanisms that either don't work very well (RLHF) or are entirely absent.
I do not usually put links to my own posts in comments, but I was so proud when I wrote this. Now I read in Helen Lewis’s post that Hobbes already came up with it in 1651... I need to do more reading in the Humanities, I’m afraid.
"The taxpayer is the hero of the modern world" is a strangely uplifting thought, isn't it? Would be nice if politicians reframed the paying of tax from the current paradigm (an unfortunate necessity to keep the show on the road) to instead see it as a noble act which, through the social contract, will help us all live better lives in the long run.
One thing I’ve become more convinced of lately is that there might very well be better political systems you could design to make sure you get the best leader possible at any one time but there is no better system than democracy for enabling the people to get rid of the duds, the dangerous and the tyrants
Maybe corporate board structures and shareholder capitalism could do the same thing too but annoyingly we’ve created the investment firm to corporate executive merry go round where the big funds have enough votes to make sure they keep voting the insiders back in
Yes, i felt something similar — oppositional politics is inherently good, because you want to have multiple perspectives articulated. Sometimes the unfashionable minority one will be right.
Oh, it can. Otter, Trint and Rev are all good (and improving all the time). There are, however, security issues — the data is sent out to servers and stored in the cloud. So I wouldn’t use it for anything sensitive or off-the-record.
AI joke: no truth in the rumour that the Geordie AI is called YI
I don't think "corporations and states are AIs" should make us *less* worried about AI existential risk. Corporations and states are slow AIs that run on humans, with well-understood alignment mechanisms (the market, elections). And they still do bad things all the time! That should IMHO make us *more* worried about fast AIs running on silicon with alignment mechanisms that either don't work very well (RLHF) or are entirely absent.
I do not usually put links to my own posts in comments, but I was so proud when I wrote this. Now I read in Helen Lewis’s post that Hobbes already came up with it in 1651... I need to do more reading in the Humanities, I’m afraid.
Still: People Are Spare Parts https://open.substack.com/pub/johnnyburger/p/people-are-spare-parts?r=1o5d4n&utm_medium=ios&utm_campaign=post
"The taxpayer is the hero of the modern world" is a strangely uplifting thought, isn't it? Would be nice if politicians reframed the paying of tax from the current paradigm (an unfortunate necessity to keep the show on the road) to instead see it as a noble act which, through the social contract, will help us all live better lives in the long run.
Some interesting background, but I fear the industrial censorship complex more than Trump or the 4 deadly issues of Silicon Valley.
Really enjoyed this interview
One thing I’ve become more convinced of lately is that there might very well be better political systems you could design to make sure you get the best leader possible at any one time but there is no better system than democracy for enabling the people to get rid of the duds, the dangerous and the tyrants
Maybe corporate board structures and shareholder capitalism could do the same thing too but annoyingly we’ve created the investment firm to corporate executive merry go round where the big funds have enough votes to make sure they keep voting the insiders back in
Yes, i felt something similar — oppositional politics is inherently good, because you want to have multiple perspectives articulated. Sometimes the unfashionable minority one will be right.
If AI can speedily produce transcripts of interviews, skeptical journos are gonna quickly get in line!
Oh, it can. Otter, Trint and Rev are all good (and improving all the time). There are, however, security issues — the data is sent out to servers and stored in the cloud. So I wouldn’t use it for anything sensitive or off-the-record.
Could it learn to read my shorthand, I wonder? O brave new world.