How a US/China Superintelligence Arms Race Will Play Out - Leopold Aschenbrenner
ฝัง
- เผยแพร่เมื่อ 5 มิ.ย. 2024
- Full Episode: • Leopold Aschenbrenner ...
Transcript: www.dwarkeshpatel.com/p/leopo...
Apple Podcasts: podcasts.apple.com/us/podcast...
Spotify: open.spotify.com/episode/5NQF...
Read the new essay series from Leopold this episode is based on here: situational-awareness.ai/
Follow me on Twitter: / dwarkesh_sp
Follow Leopold on Twitter: x.com/leopoldasch - วิทยาศาสตร์และเทคโนโลยี
a drinking game of everytime the guest says "you know"
That would be a fatal game to play
@@kroyno1 asked gemini, apparently it was mentioned 72 times in 7 minutes :D
I hadn’t noticed but now that’s all I can focus on
It's very American
There is the "like" combined with "like you know"
Slow it down to 0.75x and the guy just sounds like one of those people who get drunk and talkative.
Omg ! That is actually true and its funny you figured that one out…
And I thought Dwarkesh was intense sometimes....
This guy’s Situational Awareness article reminds me of the Coronavirus post on Medium by Tomas Pueyo back in March 2020. A strong take on a critical, fast-moving issue. And a take that is useful as something to argue against and figure out your personal position on many matters.
Thank you. This helps me more than you know 🙏
think this guy's predictions are maybe just a little bit hyperbolic lol
You do not think AI is going hyperbolic?
@@MetaverseAdventures I mean sure, but he's talking about things on the scale of trillions of dollars of investment but also like they're sudden and unsurprising. Nothing that takes pushing double digits percents the US GDP is gonna be just here all of the sudden before we know it lol. (If it even happens at all lol. The infrastructure for that much electricity would take many many years to create.) Let alone the fact that this supposed trillion dollar training run is like 3 or 4 generations of training runs in the future.
Also, say we do get a superintelligent AI all the sudden, he makes it sound like the world will instantly change, there will be new WMD's all the sudden etc. I mean sure it will change fast, but we won't have crazy new military capabilities on the order of 3-6 months like he's talking about. Sure maybe the NSA will get good and we can hack infrastructure well all the sudden. But software people *always* underestimate implementing things in the real world lol. Say a superintelligence designs a new generation of hypersonic glide vehicles that is a game changer or something like that, actually funding, creating a supply chain, producing, etc... all that stuff takes years.
I don't really doubt it's coming, and I don't really doubt things will change quickly. I just think he's way over the top about it.
@@CodyRay295 I don't get it as he never said anything explicitly about the time frame just the tends and for sure not 3-6 months. It is hyperbolic be it 2-3 years or 10-20. Either way, it is going to be an insane time to live.
@@CodyRay295 These processes are only slow now because there's no need for them to move faster - the current cost is not justified. But as Leopold rightly mentions, WWII demonstrated a time when the strategic need emerged for rapid progress, and look how fast the US was able to develop and deploy brand new military technologies, including WMDs like the atomic bomb. If the geopolitical race to AI-driven military supremacy does heat up, then the strategic need for faster development and deployment cycles will emerge - not to mention the cost of such acceleration will continue to decrease as these processes are further automated. Eventually it's obvious that all of the physical and virtual processes involved in all of these supply chains will become full-automated with AI management and robotic labour. That will enable the acceleration, and any close geopolitical competition will require it.
@@AngusTatchell okay so when China invades Taiwan I guess we will see if our priority is spending most of our GDP on AI training or actual supply chain logistics.
Kurtan has grown up so much since filming This Country! Bless
Has a bit of a lunatic vide about him this guy, but interesting scenarios to ponder even if it ends up all being just fanfiction
Cringe
Be silly to think the gov wont millitirise it, if they havent already. Doubt its fanfic.
Its interesting when we debate things like what the military industrial complex, the private super powers and what the ccp should do to make the world a better place. Those institutions tend not to put things up for a vote.
Indeed, and both private institutions alongside the military industrial complex have far too much influence in our democratic institutions
Jihad like simple minded religion around SAI and Scale Law has gone too far. The world and innovation are not that simple.
How much the Host and the guest understand what they are speaking.
This interview probably marks the start of the B2G AI industry
Some people are smart enough to make coherent and convincing arguments that are completely unrealistic.
Technologists consistently overestimate the future adoption of technology.
Overestimate some, underestimate others, completely miss others. They don't have any more special insight into the adoption or expansion of new technologies than historians or political analysts.
Ray Kurzweil's a good example. He based his predictions of AI purely on increasing computing power, and made many wildly off-base predictions that didn't come close to coming true. But he also completely missed the transformer breakthrough of 2017 and that software would progress so fast, such that now his predictions now don't look so crazy. So he was right for the wrong reasons.
He writes on his iPhone connected to satellites across the entire world in a modern healthcare system with advanced machinery all around. 😂😂
@@squamish4244 well to some extent he was actually right for the *right* reasons - the transformer breakthrough (among other algorithmic breakthroughs) closed the gap between computing power and AI capabilities (closing that gap early was OpenAI's stated reason for releasing ChatGPT when they did - so society could adjust to the scale of compute growth and resulting capabilities growth). If anything, Kurzweil was only wrong on a shorter time scale due to what was effectively statistical noise resulting from a period of unexpectedly slow algorithmic progress, but his longer-term predictions may not be that far off anymore.
@@beelikehoney Ray was making predictions like computers would have disappeared into our clothing by 2009. And that everyone would be living to 120 by now, which they obviously aren't. I lost three aunts and an uncle in the last four years at 62, 69, 70 and 70. Where's LEV, Ray?
@@AngusTatchell His raw compute predictions are way off. For those to be right, we'll need an incredible increase by 2029.
Ray's 100 pills a day regimen and "I reprogrammed my biology" schitck is bunk. I saw him on Lex Fridman two years ago and then on Joe Rogan a few months ago and holy crap has he slowed down. He was so lethargic I thought he might be ill. In a talk with Geoffrey Hinton, Hinton was running rings around him.
Fun fact, if you over saturate a market the supply of supply and demand goes up, and if your placing your bets on something that is improving it can be beneficial. I don't think China knows that type of calculation yet
Would you elaborate on this please
Open ai should stop focusing on ai development and should start focusing on security.
Leopold is very much on point and those making jabs are just projecting their own fears. I suspect he got fired as like many in the Spectrum ("assuming" as I can relate heavily to him) we say things that make people uncomfortable but are true nonetheless even if spoken "out of line". He is likely accurate in his statements as he is not making up possible futures from pure imagination but rather is looking at deep historic trends, and extrapolating recent AI growth data into the not too distant future. Sure things might and likely will unfold in a different way, but the overall trajectory is rather clear. It is a warning that we should ALL be paying attention too as there is a war for energy coming and it will impact everyone. I fear humanity more than I fear super intelligence.
Leopold seems somewhat naïve regarding reigning-in China.
So how do human workers displaced by drop in agents feed their families?
They will need to own agents that feed them.
From the Soylent Green produced by other workers
Didn't Moore's Law for All propose UBI for everyone...
It's... like, ... funny... like... sometimes I don't ... like ... know like... what I'm talking about. You know?
I am 1:45 min into the video and still have no idea what this guy is trying to talk about
I can see why they fired him
High intelligence plus low life experience and maturity can create volatile flares. His time horizon needs to catch up with everything else and balance out.
Even though I agree with most of the statements, I still find this to be entirely unconvincing...
The main problem is that the argumentative connections come across as cherry-picked. As in: Sure, the gulf war and WW2 examples might support the case, but some other examples might point into a completely different direction.
Also, while the overall argument of there being "some kind of AI arms race between the USA and China" is essentially a given, the real question is about the timescale: Is it really going to be about months? Or is it going to be years? Or decades? This makes a huge difference. If it is the latter, then it isn't really substantially different from previous arms races.
So basically, we don't really know anything...
If we act as if its months and it ends up being years we are ready. If we act as if its years and its months we are ruined
@@jaylewis9876that's assuming the point for the CCP is to ruin the US. That's not obvious. Domination is more likely, but also much less scary to theorize about.
@@d000dez Domination is much easier if the other country is in ruins
@@d000dez you need to read up on the realist theories of IR, which seem to motivate China, Russia and the authoritarians much more than any other (Mearsheimer even says they've admitted as much to him). The CCP's goal, along with any rising rival power, is going to be to destroy their rivals before they do the reverse. The US goal is clearly to contain China and prevent their rise before they risk being unseated as global hegemon. So long as the CCP holds power in China, China will be a rival to the West and democratic/free world. And so long as that rivalry exists, there is a structural / systemic motivation for China to try to dethrone the US and the West (to guarantee its own survival). It is the same logic that motivates the proliferation of nuclear weapons in rivalrous states like Iran or North Korea. It's quite simple logic really. It isn't at all convincing that the CCP would suddenly competely change its philosophy to the West, unless the West were to suddenly completely change its own. The dice is cast. Game theory is in the driving seat here.
@@d000dez Tibet and Hong Kong can explain how well CCP dominance has gone so far
Guy is an AGI
Btw, bro is so handsome.
I know right, he's gorgeous
Leopold describes the dangers of an AI arms race that many many MANY others have also stated that not only could happen, but is happening (people were just fired from OpenAi for trying to share info with China), and the comments are all dismissive and full of mockery.
I can't understand the disconnect? What's the reasoning behind saying Leopold is way off base? I don't personally think that the "peace" offering will ever happen, but I can easily see the scenario that precedes that occurring.
Because this is TH-cam, second only to Facebook in the "quality" of commenters.
Also: do you have a source that says people were fired from OpenAI for trying to share info with China? I can't seem to find anything when searching for it
@@TheLegendaryHacker 😂 so true
@@TheLegendaryHacker The company was actually Google and it happened this past march. A googleable headline is "Ex-Google Engineer Charged With Stealing AI Secrets For Chinese Companies"
I must have seen it on the OpenAI subreddit, you'll see that entry on google too. Google is actually the AI company with the biggest compute capability right now.
Right. This is happening. He is just explaining how. We cut China off from chip tech for a reason. Nvidia stock 1200 for a reason.
we're fucked
So much spin.
This guy sounds like the tech bro version of an evangelical preaches about the rapture soon
Move to China
Smoke a doobie and chill out bro
Im there dude! 👌
The red scare...
The reality is the technology being developed is scary from both sides the situation could easily be coming out of control if it keeps evolving
@@13thbiosphere 100%! I would definitely prefer the US to be leading the race, but that being said, I am still almost as terrified of Trump having control of this tech as the CCP.
Just because their reaction was wrong back in the day, doesn't mean they shouldn't have been afraid...
If you're under the illusion that the CCP is some noble, non-totalitarian, indoctornation-averse, moral virtue seeking goverment, there's a brand new bridge in downtown SF of which you can be the proud new owner...
@@13thbiosphere sounds like what was said during the cold war. Nukes are scary, too.
Dude has never seen a gym
He's pump his brain
Casually talking in the bar about AI war and doomsday 😂😂
Why does this guy look like cgi
I love how this guy says "we", as if the so called "American people" and the other citizen slaves of the West, are actually represented by the U.S. & western governments and tech companies lmfao.
Arent we?
@@quantumspark343 😂🤣 You're joking, right?
A babbling pile of nonsense
After seeing him. I'm convinced the American education system is f*cked up 😆