While I used to work at OpenAI, all of this is based on publicly-available information, my own ideas, general field-knowledge, or SF-gossip. - Leopold Aschenbrenner
Worth reading, specifically around "Counting the OOMs," was just updated on June 6th
Thanks for sharing this. I found it very interesting. But where to from here as a pleb? I flip flop between brushing it off to a degree and it being a very real mega-innovation that changes everything.
Maybe in that sense it is like Bitcoin, if you don't "get it" you just sit on the blue pill sidelines ignorant, but those of us who get it are going deep.
What else do I need to read or think about to "get" AI?
reply
10 sats \ 6 replies \ @Car OP 3 Jul
maybe @cascdr can jump in on this
reply
There's a great book called AI Superpowers worth checking out. It's written from the perspective of the guy that ran Google China and who now actively invests in chinese startups.
He covers a similar statist angle of US vs China. Imo he talks his book but has interesting perspectives chiefly that the simple fact of there being way more people in China gives certain edges (way more data for training) meanwhile the American tendency to innovate and ignore authority has its own advantages.
I'll need time to pore over the attached pdf. It can be hard to know what to believe especially in a world where so many people are talking their book and creating doom & gloom simply so they can get regulatory moats. Regardless of how this goes I have conviction that there are large productivity gains to be had and that it's imperative that we create censorship resistant, pro human and decentralized AI systems.
reply
Thanks. It can all be a bit overwhelming and tiresome.
On another note, something I (again) observed today on the train in Tokyo was the rather obvious but incredible addiction that people have with their smartphones.
It was like every single one of them were being sucked into the little glass screens with a passive expression and a disembodiment that can perhaps be compared to the opium craze that overtook China in the 19th century. The little ticks they had as they scrolled and touched the precious phone, laughing and talking to themselves, in any other situation they would be diagnosed with an illness.
This addiction to information has been known for sometime, but the image of literally everyone on the train being plugged into another world was very interesting and begs questions about the territorial authority in that new digital world, and the relationship with state, big corps etc.
Is AI not some continuation of this great information fetish? A machine that can know all things at once? Is that not why it is such a powerful stimulant, even if it is potentially make-believe?
reply
Yea I remember Rogan bringing up the point that if aliens came here and saw us staring into phones like we are, they would mistake the phones for the masters of the planet, not humans. If it weren't for the work I'm doing + lifestyle I'm in I would engineer my life to look at phones the bare minimum.
I've been reading the Screwtape Letters as recommended by @bitcoinplebdev on @Car 's podcast. The premise of the book is a series of letters between demons scheming on how to get a man's life off track. You could make an argument that in a sense smartphones are like demons assigned to us to take advantage of our lower, most base instincts and lead us astray.
There's a significant danger of AI being a more sophisticated form of that type of demon with the added bonus of the total subversion of your privacy. For that reason I created CASCDR which I hope will be a place where AI can serve genuine human needs. Luke & I are building it in a way that preserves privacy and fosters the decentralized, free open markets that I believe are needed to make it a force for good. If you're interested this talk I gave at @PlebLab startup day is a good primer on my vision and what CASCDR is trying to accomplish.
reply
Thanks I am going to read the Screwtape Letters. I am excited to see the growth of CASCDR and your ventures.
reply
Latest CASCDR update here: #600228
reply
deleted by author
reply
Reading this article through to the end now I think it is good stuff and insightful. The thing that stands out is the inherent statist bent to the author's call to arms for setting up a quasi-Manhattan project team to tackle AGI and ensure national security, alignment and supervision as the inevitable super intelligence emerges.
In other words: make sure the Americans get AGI before the "Chinese" or whoever.
I question what makes him think the US Govt is a more ethical and moral player though? He unironically cites Gulf War 1 (Advanced US tech vs old soviet Iraqi tech) and the A-Bomb as situations where the US used superior technology to commit unilateral murder, going on to imply that whoever gets AGI first (and specifically the ability to train AI AI-Researchers as a feedback loop) will be the first ones to the "bomb", but then seems to insinuate that it can only be the Americans who are allowed to have this tech.
He also talks about Covid and what a difference a couple of months of head start can make, even if it turns out it was the USG-affiliated labs behind the leak the whole time.
I would be keen to read another take on this topic that does not lean on the state or government to save the day, because they won't.
reply
there are actually quite a few...this is the best one in my opinion.
reply
I cant wait to get daylight reader. Thank you for sharing this.
reply
21 sats \ 1 reply \ @k00b 3 Jul
I haven’t read that yet but I linked to a great pod with him recently.
reply
It's def a long slog read. Starting to understand what the long tail to inference data is and whether "Big AI" gets to AGI by 2028.
a great pod
Cool ill check it out.
reply