Open AI Super Alignment former employee
A very shocking blog post was published by a former Open AI Super Alignment employee (Mr. Leopold-Aschenbrenner).
“There are probably only a few hundred people in the world who understand what is about to hit us right now, understand how crazy things are, and have circumstantial awareness.”
Super intelligence by 2028-29, and AGI by 2027. The reason why this was realized is that it suddenly enters a phase transition (GDP of 30% or more) and an acute risk period (rapid rise in tension between nations) of economic growth unparalleled in human history thereafter. Also, it is starkly described that human history literally changes due to the rapid emergence of intelligence that qualitatively surpasses humans from AGI (from a few months to 1 year, 2 years at the latest).
Bringing the American economy into a wartime system in order to manufacture 100 million GPUs; blocking everything, eliminating spies, and dodging an all-out attack by the Chinese Communist Party; somehow managing 0.1 billion of AGI that furiously automates AI research, makes a 10-year leap forward in 1 year, and eventually creates AI systems far smarter than the smartest humans; developing new technologies necessary to stabilize the situation and get ahead of the enemy, and the US military to integrate them It will be rapidly remodeled. While dealing with a tense international situation that you've probably never experienced before.
I feel that this article is an article that all mankind, not to mention the government, should see.
It seems that there is considerable confidence that AGI will first be achieved by simply increasing the number of digits (OoMs) due to scaling of computational resources (increase in effective computational volume after including all techniques other than prior training such as RLHF and CoT).
After that, AGI first specialized and automated AI research, and there were 0.1 billion automated researchers, each working 100 times faster than humans. It is estimated that this will happen by 2028. After that, it was generalized into all kinds of science and technology research, and automation, robotics was not a hardware issue, but simply a software issue, and was solved at an overwhelming speed in the years after AGI. We are rapidly falling into the most extreme situation humanity has ever faced.
We will talk about investing in training calculation resources on the scale of 1 trillion dollars (suggests they will be used for superintelligence training by 2030), talk about plans to use hundreds of billions of dollars in training costs by 2028, and then move on to national security.
The end of the current nuclear deterrent system due to overwhelming technical power due to superintelligence inventions (nuclear weapons and stealth submarines can be preemptively neutralized), preemptive attacks (in the extreme, attacks on AI-related facilities by special operations forces) and the possibility of war by China, which rapidly felt national security issues during the transition process to a fundamentally different world order, have been clearly asserted.
“Imagine what it would be like to experience the development of military technology in the 20th century in less than 10 years. After a few years, they would have transitioned from horses and rifles and trenches to modern tank forces, a few more years later to supersonic fighters and large armies of nuclear weapons and intercontinental ballistic missiles, and a few more years later, they would have moved to stealth and precision that could defeat enemies before they even knew it. This is the situation we face with the advent of superintelligence. A century's worth of military technology advancements will be compressed in less than a decade. Superhuman hacking, robot armies, and swarms of autonomous drones that can neutralize much of the enemy's military power will appear, but what is even more important is the invention of a completely new paradigm that we still can't even imagine, and new weapons of mass destruction weapons with destructive power increased 1000 times (and new weapons of mass destruction defenses such as impenetrable missile defense that rapidly and repeatedly upend the balance of deterrence).”
Currently, advanced companies don't care too much about national security, and security is clumsy, and when Frondia model algorithms are eavesdropped or spied on at San Francisco parties, values are fixed by authoritarian countries, and the possibility of “semi-permanent loss of liberal democracy” is seriously pointed out. It is pointed out that this state is “insanity.” It would be nice if the superiority of China and America could be secured for a few years, but in a few months, no alignment can be worked on, and humanity may be destroyed as it is. Security above the military level and handling of nuclear weapons is necessary for AGI-related matters.
Then, within just a few years, this fact will drastically repaint the American government's Overton window, and he pointed out that America becomes/should become a “wartime system.”
This might be a historic article. I experienced this 20th century geopolitical fervor and man-made crisis in just a few years. We are about to enter such an era. They seem optimistic about AI alignment, but they say there is a serious possibility that if there is no room in competition with China, it will fail and humanity will risk survival. Start an article I've personally read in my life where I don't feel this much of a sense of reality. Is this real?
Finally, regarding AGI feasibility, “A few years ago, at least for me, I took these ideas seriously, but they were abstract and confined within models and probability estimates. It feels very real now. I can see it. You can see how AGI is built. It is no longer an estimate or hypothesis of the size of the human brain, theoretical extrapolations, etc., and it is basically possible to tell the clusters where AGI is trained and when they are built, rough combinations of algorithms used, unsolved problems and the path to their solution, and a list of important people.” It's written, and you can see that it's already quite real.
This article is similar to the timeline that Daniel Kokotajlo, Sam Altman, and John Schulman often mentioned within OpenAI. In other words, it can be said that this depicts information about the future within Open AI quite clearly. Also, the reason they are optimistic about super alignment is that deep learning is easy to generalize, etc., as a reason for creating a unique section.
You can understand that they are trying to change the world and human history. This fact is an article that confronts reality from various obvious data (investment trends, algorithm improvement rates, semiconductor trends, etc.).
And it seems that there are only “a few hundred people” who recognize this reality as “SITUATIONAL AWARENESS.”
Everyone should read it.
situational-awareness.ai
“There are probably only a few hundred people in the world who understand what is about to hit us right now, understand how crazy things are, and have circumstantial awareness.”
Super intelligence by 2028-29, and AGI by 2027. The reason why this was realized is that it suddenly enters a phase transition (GDP of 30% or more) and an acute risk period (rapid rise in tension between nations) of economic growth unparalleled in human history thereafter. Also, it is starkly described that human history literally changes due to the rapid emergence of intelligence that qualitatively surpasses humans from AGI (from a few months to 1 year, 2 years at the latest).
Bringing the American economy into a wartime system in order to manufacture 100 million GPUs; blocking everything, eliminating spies, and dodging an all-out attack by the Chinese Communist Party; somehow managing 0.1 billion of AGI that furiously automates AI research, makes a 10-year leap forward in 1 year, and eventually creates AI systems far smarter than the smartest humans; developing new technologies necessary to stabilize the situation and get ahead of the enemy, and the US military to integrate them It will be rapidly remodeled. While dealing with a tense international situation that you've probably never experienced before.
I feel that this article is an article that all mankind, not to mention the government, should see.
It seems that there is considerable confidence that AGI will first be achieved by simply increasing the number of digits (OoMs) due to scaling of computational resources (increase in effective computational volume after including all techniques other than prior training such as RLHF and CoT).
After that, AGI first specialized and automated AI research, and there were 0.1 billion automated researchers, each working 100 times faster than humans. It is estimated that this will happen by 2028. After that, it was generalized into all kinds of science and technology research, and automation, robotics was not a hardware issue, but simply a software issue, and was solved at an overwhelming speed in the years after AGI. We are rapidly falling into the most extreme situation humanity has ever faced.
We will talk about investing in training calculation resources on the scale of 1 trillion dollars (suggests they will be used for superintelligence training by 2030), talk about plans to use hundreds of billions of dollars in training costs by 2028, and then move on to national security.
The end of the current nuclear deterrent system due to overwhelming technical power due to superintelligence inventions (nuclear weapons and stealth submarines can be preemptively neutralized), preemptive attacks (in the extreme, attacks on AI-related facilities by special operations forces) and the possibility of war by China, which rapidly felt national security issues during the transition process to a fundamentally different world order, have been clearly asserted.
“Imagine what it would be like to experience the development of military technology in the 20th century in less than 10 years. After a few years, they would have transitioned from horses and rifles and trenches to modern tank forces, a few more years later to supersonic fighters and large armies of nuclear weapons and intercontinental ballistic missiles, and a few more years later, they would have moved to stealth and precision that could defeat enemies before they even knew it. This is the situation we face with the advent of superintelligence. A century's worth of military technology advancements will be compressed in less than a decade. Superhuman hacking, robot armies, and swarms of autonomous drones that can neutralize much of the enemy's military power will appear, but what is even more important is the invention of a completely new paradigm that we still can't even imagine, and new weapons of mass destruction weapons with destructive power increased 1000 times (and new weapons of mass destruction defenses such as impenetrable missile defense that rapidly and repeatedly upend the balance of deterrence).”
Currently, advanced companies don't care too much about national security, and security is clumsy, and when Frondia model algorithms are eavesdropped or spied on at San Francisco parties, values are fixed by authoritarian countries, and the possibility of “semi-permanent loss of liberal democracy” is seriously pointed out. It is pointed out that this state is “insanity.” It would be nice if the superiority of China and America could be secured for a few years, but in a few months, no alignment can be worked on, and humanity may be destroyed as it is. Security above the military level and handling of nuclear weapons is necessary for AGI-related matters.
Then, within just a few years, this fact will drastically repaint the American government's Overton window, and he pointed out that America becomes/should become a “wartime system.”
This might be a historic article. I experienced this 20th century geopolitical fervor and man-made crisis in just a few years. We are about to enter such an era. They seem optimistic about AI alignment, but they say there is a serious possibility that if there is no room in competition with China, it will fail and humanity will risk survival. Start an article I've personally read in my life where I don't feel this much of a sense of reality. Is this real?
Finally, regarding AGI feasibility, “A few years ago, at least for me, I took these ideas seriously, but they were abstract and confined within models and probability estimates. It feels very real now. I can see it. You can see how AGI is built. It is no longer an estimate or hypothesis of the size of the human brain, theoretical extrapolations, etc., and it is basically possible to tell the clusters where AGI is trained and when they are built, rough combinations of algorithms used, unsolved problems and the path to their solution, and a list of important people.” It's written, and you can see that it's already quite real.
This article is similar to the timeline that Daniel Kokotajlo, Sam Altman, and John Schulman often mentioned within OpenAI. In other words, it can be said that this depicts information about the future within Open AI quite clearly. Also, the reason they are optimistic about super alignment is that deep learning is easy to generalize, etc., as a reason for creating a unique section.
You can understand that they are trying to change the world and human history. This fact is an article that confronts reality from various obvious data (investment trends, algorithm improvement rates, semiconductor trends, etc.).
And it seems that there are only “a few hundred people” who recognize this reality as “SITUATIONAL AWARENESS.”
Everyone should read it.
situational-awareness.ai
Disclaimer: Community is offered by Moomoo Technologies Inc. and is for educational purposes only.
Read more
Comment
Sign in to post a comment
KIMIHIKO OP : Incidentally, if Situational Awareness is true, it can be seen from tables and mentions that Nvidia and Microsoft stock prices will increase many times in the future. Anyone who reads this often will notice it.