朋友们又出大瓜了

朋友们又出大瓜了，就在刚刚，OpenAI吹哨人，丹尼尔·科塔洛伊，揭开了AI巨头们，不想让你知道的真相。 Yes. For reasons I'm happy to get into. But the main thing I've learned is when I go talk to people at Anthropic and OpenAI about forecasting, they're like, "It's not going to take that long. You need to shorten them again. Get them back to 2027 or 228 because these powerful CEOs, Ariel or Sam or Elon are each other to be in control of the most powerful AIs and are literally afraid that if the other guy gets there first, he might become dictator. I mean, Anthropic is on track to be the entire economy by 2030. But none of these people should be trusted with that much power. So, this is the most important thing happening in our lifetimes, probably in all of history, in fact. And it's very important that it goes well. So, I think that there's a lot we can do to steer things in a better direction. There's loads of benefits that we can get from AI if we do it right. And if we do solve the problems, then things could be absolutely amazing for everyone. Well, this report here in 2021, it was remarkably accurate. And then you just published this one. Yeah. So, this is our new scenario. So, let's go through these slowly. I mean, you know, there's a lot of people who think that this is sort of like their lives as they don't want to be uploaded. I don't think they'll ever join the workforce. Everybody should be afraid that their jobs are going to be lost. And I know this because I went to open air in 2022. What I did there was forecasting what the next couple years like. And then I've done my research and I've compiled it. I'm going to share with you guys. So, the first question is, why is AI 70% likely to lead to human extinction? The danger is not that AI will one day suddenly go on a rampage or that robots will go crazy. It's that AI is too useful. It first helps you write code, do research, analyze intelligence. We are all happy to hand over more and more work to it. So that it can enter the core of scientific research, economic and military affairs, and grasp enough real-world resources. At that time, humans want to take back power. It may be too late. The fourth and most incomprehensible point is that no one dares to stop first. OpenAI will think, if we stop, will Anthropic succeed first? If we slow down, will Google and Musk get superintelligence first? Everyone knows this is very dangerous, but everyone also feels that they absolutely cannot let others get it first. Everyone thinks, only if I win first, the world will be safe. The second question, why might superintelligence arrive around 2030? Many people think this is alarmist. Because in our experience, AI is not that powerful. But Daniel said, don't look at feelings, look at data. The number of neurons in the AI brain, that is, the parameters. In 2020, it was 175 billion. Now it's 10 trillion. In 6 years, it has grown by about 60 times, and it is still growing wildly. But what's truly alarming is that the bunch of neurons are doing one thing: letting AI develop AI. Many people think AI will replace customer service first, then replace lawyers, doctors, and programmers. But the real order might be completely reversed. AI companies now want to replace not ordinary people, but their own AI researchers. First, let AI write code itself. Second, let AI conduct experiments itself, analyze results. Third, let AI design the next generation of stronger AI. Once this cycle starts, the technical progress that originally took 3 to 5 years might be compressed to a few months or even half a year. This is the so-called recursive self-improvement, that is, intelligence explosion. What makes him more uneasy recently is that many people in other industries believe that superintelligence is unlikely to appear by 2030. But Daniel said, and many people in OpenAI and Anthropic have told him, that as more and more people tell him, it's possible that by 2030, it will be conservative. Maybe it will appear in 2028, 2029. The third question, how can humans save themselves? After Daniel left OpenAI, he spent another year writing a more optimistic plan, the "AI 2040" plan. This is what he sincerely hopes we will walk. He believes the core is not to completely eliminate AI, but to push superintelligence from around 2030 to 2040, using these ten years to solve safety supervision and distribution problems, and exchange for human survival opportunities. So, how to do it specifically? Four points. First, slow down through supervision and international agreements, and let the development of AI fall to a speed that humans can understand and control. Second, transparency, which requires training new models with recipes, architectures, and research results, and cannot be locked in a black box by a few companies. Instead, it should be seen by more scientists and regulators. Third, sharing. Superintelligence cannot belong to a company, a country, or a few people. Let multiple countries and multiple companies have relatively close capabilities. Fourth, reversibility. Daniel strongly suggests that future new data centers must be designed to be able to "one-click destruction". Once the situation is found to be out of control, humans still have the last chance to step on the brakes. In this plan, it is also proposed a plan called "Citizen's Dividend". The core meaning is: if future companies use AI and robots to replace a large number of employees, then through taxation and licensing systems, the wealth created by the AI economy will be distributed to every ordinary person. According to the ideal projection, citizen's dividend may initially be 25,000 US dollars per person per year. With the explosive growth of the AI economy, it may eventually reach 10 million US dollars per person per year. The host also asked us a question: In the next 10 years, which jobs do you think can truly surpass AI? What should children learn now so they won't be eliminated? His answer was very counter-intuitive. He said if superintelligence really appears, then almost nothing you learn will be useful. This sentence, I understand, don't manage the future after superintelligence appears. At least tomorrow, what replaces you is not AI, but colleagues who are better at using AI. So the strategy is very clear. Children must maintain their curiosity, exploration, and imagination, and add a proper amount of contact and use of AI. So we ourselves need to think as much as possible about one thing: how to use more and stronger AI, even top AI, in work. I will share more about useful AI and usage experience later. One principle: don't let it think for you, let it force you to think. It's very anti-human, isn't it? Only in this way is there a chance. This is the end of this issue, see you next issue. AI whistleblower, the truth about OpenAI whistleblower, AI has a 70% chance of destroying humanity.