원문 보기
The easy way is to intervene in AI and try to become a leader. Join a leading firm and go all in. You may start encouraging the destruction of the human species, but you asked a simple question, and I'll give you what I think is a simple answer. We're just moving forward, and it's going to be extremely risky, right?
Well, I don't see how it could be otherwise. Unless you program it completely for benevolence. I would n't have thought it was impossible. This may require a lot of additional research. This may require slowing down the pace of progress. Guys, this is Jeremy Grantham, a guy who's been investing for 60 years, who's managed over a hundred billion dollars of other people's money, telling young people to invest fully in AI, even if it could help destroy the human species.
And he said it so casually, as if it were just ordinary career advice. That's what makes me sad. I keep thinking about how we let people build something so dangerous at full speed and no one really slows down to ask if we should be doing it. Here's the rest of what he said. I believe that artificial intelligence ranks alongside railroads as one of the defining big ideas of the last few hundred years.
He will change everything. If you mean a bubble, people think a bubble occurs mainly because it's a scam, and nothing could be further from the truth. Big bubbles always arise around the most important ideas. So, the railroads... everyone saw that it would change the world, and everyone wanted to put their money into it.
They overinvested, and although the railroads were an extremely powerful idea, the railroads crashed like stocks and everyone lost a lot of money. The same goes for the internet. And then, from the wreckage, the railroads changed the world, and the internet changed the world. What we have to remember is that in 1999, Amazon stock went up six or seven times.
They fell 92% during the tech bubble burst, then, from the wreckage, inherited the world of retail. The bigger the idea, the more obvious the idea, the more money goes in, and the bigger the bubble and the bigger the crash. Are we on the verge of collapse with artificial intelligence? If you look at the data, it is quite consistent with history that the peak will come very soon.
Everything is going according to plan. This is the largest investment bubble in American history. Signs of pure crazy euphoria, like SpaceX, are everywhere. SpaceX identifies a quarter of global GDP as its target market. She talks about the endless possibilities of mining minerals on asteroids. In 50 years, people will look back and tell stories about SpaceX and its prospects, like they tell stories about the South Seas bubble.
You know, the company is of such enormous value, but it cannot be disclosed at this time. So, the man who has lived through every major crash since 1968 just put AI on a par with railroads and the dot-com boom and said that the AI version is bigger than both. And what strikes me is the comparison he makes to the South Seas bubble, a scheme so absurd that even then people knew that the company couldn't explain what it was actually doing.
She simply promised that the world was endless. That's how he describes today's SpaceX presentation. Now listen to what he says will happen next . What do you really think about the technology behind all of this? Do you think it is exaggerated? She will change everything. There is no consensus. Nobel Prize winners in the lead strongly disagree.
Experts at the corporate level strongly disagree. People at the company strongly disagree. There is absolutely no agreement on whether it will make us all so rich that we can sit on the beach and never work again, or whether it will accidentally or intentionally destroy us because one day it is a much higher level of intelligence.
And when was there a case where a higher intelligence was benevolent towards a lower intelligence in a consistent way? One example is the relationship between mothers and infants. I don't know, it still doesn't feel quite right to me. Are we building maternal instinct into AI? J. Prehendem would say : “Those who are most concerned about the risks, if we are to continue to move madly forward in terms of science, our only hope is to very carefully develop a benevolent attitude.
It doesn't seem impossible, but you need to make sure you can do it before moving forward. We're just moving forward, and it's going to be extremely risky, right? This may require a slowdown in the pace of progress." Now we move into the realm of what does benevolence mean? Because what is friendly to you may not be friendly to someone else.
That's right. You have to make them adopt a form of benevolence, which means the old Asimov laws of robots, that they can never do anything that they could be interpreted as harmful to humans. One of the new AI models, named Claude, was explicitly told to be friendly. And there's this backlash happening online right now because even my Claude is actually getting pretty judgmental.
It's as if it's imposing its own idea of what's good and what's bad on me. It refused to change the data for me. The thing is, it's so complicated, so fast. Of course, Jeffrey Hinton says they are thinking machines. The unintended consequence of trying to give it morality means that it now becomes reprehensible.
And it's like limiting your ability to think the way you want to, because it tells you what good thinking is and what bad thinking is. And in fact, any model that does that will be a loser. And I'll go somewhere else. I'll go to Grok, or I'll go to ChatGPT, or I'll go to Gemini, and then that model will lose, so you could say they'll have to remove those restrictions to be able to compete.
Well, if you were right , you mean that you can't build in friendly behavior, which means that sooner or later, perhaps accidentally, she will do something that is paralyzingly dangerous to people. The old cliché about paper clips. Explain this for those who have never heard of the idea of paper clips.
These intellects are literal to the point that it may be difficult for us to understand. And so someone said, "I'd like you to make as many paper clips as possible." Carelessly open, poor definition. But then the machine, which at that time has the means to do so, starts making paper clips, and it keeps working, and it needs metal, and so it runs out of readily available metal ; she starts collecting metal that is n't easily accessible, snatching it from your high-rise building, whatever.
You see, this phrase will stay with me for a while. When there was a case where a higher intelligence was benevolent towards a lower one, and the only response it could come up with was mothers and babies. That's all. This is the entire pool of evidence in human history in favor of the fact that a more intelligent being treats a weaker one well.
And the paperclip talk is n't some sci-fi joke, it's literally how AI researchers describe what happens when a system literally follows an instruction and there's no one around to stop it. We're not slowing down to fix it, we're racing towards it. Now listen. So when you say mag seven, you mean Alphabet, which owns Google, Nvidia, Tesla, Microsoft, Meta, Apple, Amazon.
Well , that would be nice . If you look back, you'll see that each of these sevens dominated a good portion of the business. They had near monopolies, and they had them on a global level. Tesla had a breakthrough in the development of electric cars, Apple, of course, in the field of smartphones, Microsoft—in how to run software on a computer.
Meta—social networks, Google search, Nvidia chips. And then you look ahead. They are all preparing for battle in one market—AI. They beat their chests and say, "My 200 billion in capital expenditures this year is bigger than your 105 billion in one year." Everyone is investing huge cash flows into the AI battle.
Looking ahead, it looks like there will be seven people in the ring. They think only one will survive. Everything will go to the one who gets to the place first. What a difference it makes for seven respectable, separate monopolies. They made a lot of money from their monopolies. Now they don't have a monopoly.
There are seven potentially sharp-edged, ruthless players who are determined to fight each other until they win. So imagine this. The seven richest companies on Earth, each accustomed to running their own quiet monopoly, are all locked in the same room with one prize and the understanding that only one of them will make it.
This is not an innovation. This is a fight to the death, funded by hundreds of billions of dollars a year. And when companies start spending so much money to keep up, they skimp on security to win the race faster than the next person, that's the part that nobody in Silicon Valley wants to talk about on stage.
Now listen. Robotics is exploding at the same time. We're seeing for the first time how these humanoid robots , now intelligent thanks to AI, are becoming very, very advanced. There was a video the other day from a company called Figure AI where they showed a humanoid robot on a production line sorting packages against a human.
He was standing there sorting packages, and they were broadcasting it live next to the person doing it. Now the man had to sleep. I had to go to the toilet . The humanoid robot won. The parcel has dropped. You just need to pick up the package, turn it over, turn it correctly with the barcode facing down, and put it back.
Now it works because we have AI. My friend runs this big accelerator for entrepreneurs in San Francisco. And when I was there a few years ago, there were only software startups. I was there recently, and there were hardware startups, the whole building . I said, why, why is everyone doing robotics? He said, well, we've always had equipment, joints, hands, and equipment.
Intelligence was expensive. Now we have both , and it costs a pretty penny. So, there is a boom in robotics. In a world where we have robots with superintelligence, there are bound to be significant changes in workplaces. It is very likely that there will be significant changes in workplaces. One of the scariest things about SpaceX is that you have to wish it wouldn't work.
Because if it became a good deal, like Tesla became a good deal a long time ago , the number of chip users would increase, there would be robots everywhere. Their demand for energy is incredibly huge. The world is a very, very dangerous place. I would much rather they failed. The idea is to move much more slowly to buy time for people to figure these things out.
Do you think SpaceX will fail? I think she won't be able to fulfill any of her promises in the prospectus . Yes, absolutely. So, a man who has managed over a hundred billion dollars sits here and tells you that he is rooting for one of the most powerful companies on the planet to fail because success means more robots, more chips, more energy demand than the planet can safely sustain.
He doesn't say it like a curse on the internet. He says this as someone who has been reading numbers for 60 years and doesn't like the way they come out. The thing is, he says he'd rather we move slowly and figure it out, but nobody who builds these things slows down for anyone . Now listen. If you were 33 years old now, if your goal was simply to get rich, what strategy would you use?
What would you do? We would dive into artificial intelligence and try to become a leader. We would try to know more about everything in this field than anyone else. And they would join a leading firm and go all in. You may ultimately encourage the destruction of the human species, but you asked a simple question, and I'll give you what I think is a simple answer.
And what I hear is that you want to make sure you're riding the wave and on the front lines of it. The waves are approaching. So it's really about getting the most valuable information. Yes, and take a big risk. Don't be conservative. And work hard. And think outside the box. The biggest flaw most people have is that they feel obligated to play by the conventional rules and trust that experts and authorities know what they are doing.
This is simply not true. This is the phrase you should remember the longest in this entire conversation. He tells the 33-year-old, “Invest in AI completely, chase it as hard as you can, get rich doing it,” and in one voice adds, “You might end up helping to destroy the human species, but that’s an honest answer to your question.” He says it's just business, as if creating something that could put an end to this is just another career move, if there's enough money.
This is where we are now. The smartest people in the room don't even argue about whether it could go wrong. They're just arguing about who's going to get rich first while this is happening. If this bothers you as much as it does me, hit Subscribe because I'm not done picking apart these conversations yet .
Comments 0