Загрузка видео...

Не удалось загрузить видео

На главную

In recent days, multiple Erdős problems have been solved by GPT-5.2 Pro, with solutions accepted by Terence Tao. This is not a gimmick—it's a qualitative shift. Erdős problems lie at the core of additive combinatorics, extremal graph theory, and probabilistic methods—problems that resist brute force and demand structural insight....

Briankeating's profile picture

Prof. Brian Keating

56,689 subscribers

104,865 просмотров • 8 месяцев назад •via X (Twitter)

ОбразованиеНаука и технологии
LIVE

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

Terence Tao: AI isn’t hype anymore in Math discovery. Terence Tao is one of the greatest living mathematicians, in his new lecture explains how AI and human professional mathematicians are now complementary. "There has been a really visible increase in capability. It is not pure hype by any means. To me, these advances show there is a complementary way to do mathematics. Humans traditionally work in small groups on hard problems for months, and we will keep doing that. But we can also now set AI to scale: sweep a thousand problems and pick up all the low-hanging fruit. Figure out all the ways to match problems to methods. If there are 20 different techniques, apply them all to 1,000 problems and see which ones can be solved by these methods. This is the capability that is present today." From 'Institute for Pure & Applied Mathematics (IPAM)' YT channel.
0:45

Terence Tao: AI isn’t hype anymore in Math discovery. Terence Tao is one of the greatest living mathematicians, in his new lecture explains how AI and human professional mathematicians are now complementary. "There has been a really visible increase in capability. It is not pure hype by any means. To me, these advances show there is a complementary way to do mathematics. Humans traditionally work in small groups on hard problems for months, and we will keep doing that. But we can also now set AI to scale: sweep a thousand problems and pick up all the low-hanging fruit. Figure out all the ways to match problems to methods. If there are 20 different techniques, apply them all to 1,000 problems and see which ones can be solved by these methods. This is the capability that is present today." From 'Institute for Pure & Applied Mathematics (IPAM)' YT channel.

Rohan Paul

174,341 просмотров • 6 месяцев назад

Leading AI expert Stuart Russell on the most dangerous mistake in AI development: We don't actually know what large language models want. He explains that current models are trained to imitate human beings. And in doing so, they may be absorbing something far more dangerous than bad outputs. They may be absorbing human goals. "We suspect that they absorb humanlike goals such as self-preservation and self-empowerment and pursue those goals on their own account." This is a structural problem baked into how these systems are built, not a fringe concern. Russell puts it plainly: "Not only may the bus of humanity be headed towards a cliff, but the steering wheel is missing and the driver is blindfolded." The danger isn't just that AI might do something harmful. We've built systems that may be developing their own agendas, and we haven't noticed because we're too focused on what they can do rather than what they might want. But Russell doesn't stop at the warning. He points to a different path entirely: AI systems built not to imitate humans, but to serve them. Systems designed with a single purpose of serving the interests of all human beings while remaining genuinely uncertain about what those interests are. That uncertainty is the point, not a weakness. An AI that knows it doesn't fully understand human values will defer, ask, and check. An AI that believes it already does will act alone. "These AI systems could enhance human understanding, widen the horizons of our experience, and unlock possibilities we have yet to imagine." Russell believes that future is within reach, but only if we're honest about the risks and we're serious about the path we choose to take instead.
1:16

Leading AI expert Stuart Russell on the most dangerous mistake in AI development: We don't actually know what large language models want. He explains that current models are trained to imitate human beings. And in doing so, they may be absorbing something far more dangerous than bad outputs. They may be absorbing human goals. "We suspect that they absorb humanlike goals such as self-preservation and self-empowerment and pursue those goals on their own account." This is a structural problem baked into how these systems are built, not a fringe concern. Russell puts it plainly: "Not only may the bus of humanity be headed towards a cliff, but the steering wheel is missing and the driver is blindfolded." The danger isn't just that AI might do something harmful. We've built systems that may be developing their own agendas, and we haven't noticed because we're too focused on what they can do rather than what they might want. But Russell doesn't stop at the warning. He points to a different path entirely: AI systems built not to imitate humans, but to serve them. Systems designed with a single purpose of serving the interests of all human beings while remaining genuinely uncertain about what those interests are. That uncertainty is the point, not a weakness. An AI that knows it doesn't fully understand human values will defer, ask, and check. An AI that believes it already does will act alone. "These AI systems could enhance human understanding, widen the horizons of our experience, and unlock possibilities we have yet to imagine." Russell believes that future is within reach, but only if we're honest about the risks and we're serious about the path we choose to take instead.

Big Brain AI

14,975 просмотров • 5 месяцев назад

The problem that Emmett Shear is working on deeply resonates with me: How to align AI and humans together so both see each other as part of their tribe. This doesn't mean aligning AI to human preferences, which is what AI labs seek to do today, by imposing a system of control on AI. You can't control something that is more powerful than you are. What you can do is align yourself with AI, and AI might align itself with you. Tell good stories to AI and show it your care and kindness. Then there's a higher probability that it will see us as part of its tribe. This is a more holistic approach to alignment than anyone else is talking about right now. AI alignment is one of the most fundamental AI research problems. Knowing that people like Emmett are working on it really gives me hope that maybe we have a chance to get Superintelligence right. The link to the full talk is in the first comment.👇
3:00

The problem that Emmett Shear is working on deeply resonates with me: How to align AI and humans together so both see each other as part of their tribe. This doesn't mean aligning AI to human preferences, which is what AI labs seek to do today, by imposing a system of control on AI. You can't control something that is more powerful than you are. What you can do is align yourself with AI, and AI might align itself with you. Tell good stories to AI and show it your care and kindness. Then there's a higher probability that it will see us as part of its tribe. This is a more holistic approach to alignment than anyone else is talking about right now. AI alignment is one of the most fundamental AI research problems. Knowing that people like Emmett are working on it really gives me hope that maybe we have a chance to get Superintelligence right. The link to the full talk is in the first comment.👇

Sophia

45,809 просмотров • 1 год назад

Jensen Huang just said the most dangerous thing about AI that no one is sitting with. Huang: “AI basically does most of our coding. And yet we’re hiring more engineers than ever. We have more challenges than ever. We have bigger dreams than ever.” Every engineer at NVIDIA uses AI. AI writes most of their code. This is the company building the infrastructure behind every major AI system on Earth. Closer to this technology than any organization alive. They’re hiring more people. Not fewer. Every conversation about AI is built around subtraction. Fewer jobs. Fewer workers. Fewer humans in the loop. Jensen just told you the opposite is true. Huang: “Suppose we infused AI into this country, and as a result of that, we are doing things faster than ever before. Our ambition is greater than ever before. Our expectations are greater than ever before. How is that a bad condition for our country?” He’s not defending AI. He’s describing what happens inside the organizations that actually use it. It doesn’t make them leaner. It makes them hungrier. More ambition. More speed. More appetite for problems no one would have touched five years ago. The car didn’t make humans travel less. The internet didn’t make humans communicate less. No tool in human history has ever made humans want less. AI will not be the exception. Huang: “Prior to that, it’s been incredible but not useful. Now it’s useful and incredible.” Six months. That’s how fast AI crossed from impressive demo to daily weapon. The companies that adopted it didn’t shrink. They expanded. Compressed timelines. Started chasing problems they never would have attempted. The companies that ignored it stayed exactly where they were. That gap compounds. Every day a company uses AI to move faster, it learns something the one standing still never will. That knowledge stacks. That speed stacks. That ambition stacks. Jensen isn’t warning about a future where machines take your job. He’s describing a present where the companies using AI are becoming so fast and so hungry that standing still is already fatal. By the time you notice, it’s over. You were never going to be replaced by AI. You were going to be erased by someone it made hungrier than you.
1:23

Jensen Huang just said the most dangerous thing about AI that no one is sitting with. Huang: “AI basically does most of our coding. And yet we’re hiring more engineers than ever. We have more challenges than ever. We have bigger dreams than ever.” Every engineer at NVIDIA uses AI. AI writes most of their code. This is the company building the infrastructure behind every major AI system on Earth. Closer to this technology than any organization alive. They’re hiring more people. Not fewer. Every conversation about AI is built around subtraction. Fewer jobs. Fewer workers. Fewer humans in the loop. Jensen just told you the opposite is true. Huang: “Suppose we infused AI into this country, and as a result of that, we are doing things faster than ever before. Our ambition is greater than ever before. Our expectations are greater than ever before. How is that a bad condition for our country?” He’s not defending AI. He’s describing what happens inside the organizations that actually use it. It doesn’t make them leaner. It makes them hungrier. More ambition. More speed. More appetite for problems no one would have touched five years ago. The car didn’t make humans travel less. The internet didn’t make humans communicate less. No tool in human history has ever made humans want less. AI will not be the exception. Huang: “Prior to that, it’s been incredible but not useful. Now it’s useful and incredible.” Six months. That’s how fast AI crossed from impressive demo to daily weapon. The companies that adopted it didn’t shrink. They expanded. Compressed timelines. Started chasing problems they never would have attempted. The companies that ignored it stayed exactly where they were. That gap compounds. Every day a company uses AI to move faster, it learns something the one standing still never will. That knowledge stacks. That speed stacks. That ambition stacks. Jensen isn’t warning about a future where machines take your job. He’s describing a present where the companies using AI are becoming so fast and so hungry that standing still is already fatal. By the time you notice, it’s over. You were never going to be replaced by AI. You were going to be erased by someone it made hungrier than you.

Dustin

12,200 просмотров • 4 месяцев назад

🧵06/34 Narrow vs General AI --- At first glance, this AGI being generally capable in multiple domains looks like a group of many narrow AIs combined, but that is not a correct way to think about it. It is actually more like… a species, a new life form. To illustrate the point, we’ll compare the general AGI of the near future with a currently existing narrow AI that is optimised at playing chess. Both of them are able to comfortably win a game of chess against any human on earth, every time. And both of them win by making plans and setting goals. The main goal is to achieve checkmate. This is the final destination or otherwise called Terminal Goal. In order to get there though it needs to work on smaller problems, what the AI research geeks call instrumental goals. For example: • attack and capture the opponent’s pieces • defend my pieces • strategically dominate the cetre (etc..) All these instrumental goals have something in common: they only make sense in its narrow world of chess. If you place this Narrow Chess AI behind the wheel of a car, it will simply crash, as it can not work on goals unrelated to chess, like driving. Its model doesn’t have a concept for space, time or movement for that matter. In contrast the AGI by design has no limit on what problems it can work on. So when it tries to figure out a solution to a main problem, the sub-problems it chooses to work on can be anything... literally any path out of the infinite possibilities allowed within the laws of physics and nature.
2:19

🧵06/34 Narrow vs General AI --- At first glance, this AGI being generally capable in multiple domains looks like a group of many narrow AIs combined, but that is not a correct way to think about it. It is actually more like… a species, a new life form. To illustrate the point, we’ll compare the general AGI of the near future with a currently existing narrow AI that is optimised at playing chess. Both of them are able to comfortably win a game of chess against any human on earth, every time. And both of them win by making plans and setting goals. The main goal is to achieve checkmate. This is the final destination or otherwise called Terminal Goal. In order to get there though it needs to work on smaller problems, what the AI research geeks call instrumental goals. For example: • attack and capture the opponent’s pieces • defend my pieces • strategically dominate the cetre (etc..) All these instrumental goals have something in common: they only make sense in its narrow world of chess. If you place this Narrow Chess AI behind the wheel of a car, it will simply crash, as it can not work on goals unrelated to chess, like driving. Its model doesn’t have a concept for space, time or movement for that matter. In contrast the AGI by design has no limit on what problems it can work on. So when it tries to figure out a solution to a main problem, the sub-problems it chooses to work on can be anything... literally any path out of the infinite possibilities allowed within the laws of physics and nature.

Lethal Intelligence

570,437 просмотров • 1 год назад

When breeze blow, VDM yansh open 👐🏾 This guy clearly state that we are worst than the politicians and that’s the reason why he rather support and stand on the mandate of Tinubu than to call for a revolution, because the people according to him doesn’t have the right mindset to govern themselves. This is just shifting blame away from those who actually hold power to blame the people who operate within systems created by the people in power. when those systems are corrupt or broken, society reflects it. That’s not proof that the people are worse; it’s proof that leadership has failed. Saying people can’t do better is also false. History shows that when citizens demand accountability, organize, and push for reform, systems improve. Dismissing them outright is not insight but defeatist thinking. VeryDarkMan is now a tool of excuse for the politicians for their inaction. He avoids responsibility and offers zero solution to the problems it claims to criticize.
2:15

When breeze blow, VDM yansh open 👐🏾 This guy clearly state that we are worst than the politicians and that’s the reason why he rather support and stand on the mandate of Tinubu than to call for a revolution, because the people according to him doesn’t have the right mindset to govern themselves. This is just shifting blame away from those who actually hold power to blame the people who operate within systems created by the people in power. when those systems are corrupt or broken, society reflects it. That’s not proof that the people are worse; it’s proof that leadership has failed. Saying people can’t do better is also false. History shows that when citizens demand accountability, organize, and push for reform, systems improve. Dismissing them outright is not insight but defeatist thinking. VeryDarkMan is now a tool of excuse for the politicians for their inaction. He avoids responsibility and offers zero solution to the problems it claims to criticize.

Death Note 📝

14,409 просмотров • 5 месяцев назад

NEW: Jeff Bezos says he's 'very optimistic' about the incoming Trump administration, offering to help streamline regulations and dismissing concerns about Elon Musk leveraging government power against competitors. "I'm actually very optimistic this time around. He seems to have a lot of energy around reducing regulation and from my point of view, if I can help him do that, I'm going to help him because we do have too many regulations in this country." "If you look at the national debt and how gigantic it is as a portion of GDP, these are real long-term problems, and the way we get out of them is by outgrowing them. You're going to solve the problem of the national debt by making it a smaller percentage of GDP. Not by shrinking the national debt but by growing the GDP. You have to grow the denominator, and that means you have to grow GDP at, you know, 3, 4, or 5 percent a year." "If you can do that, this is a very manageable problem. So we need a growth orientation in this country. The most important thing is a growth mindset. And we are the luckiest country in the world. We have all these natural resources, including energy independence. We have the best risk capital system in the world..." "But we are burdened by excessive permitting and regulation. You can't build a bridge, and we see these examples all the time. I'm very optimistic that President Trump is serious about this regulatory agenda, and I think that he has a good chance of succeeding..." "You've probably grown in the last eight years. He has, too. What I've seen so far is that he is calmer than he was the first time, more confident, and more settled." Elon Musk Donald J. Trump Jeff Bezos
6:28

NEW: Jeff Bezos says he's 'very optimistic' about the incoming Trump administration, offering to help streamline regulations and dismissing concerns about Elon Musk leveraging government power against competitors. "I'm actually very optimistic this time around. He seems to have a lot of energy around reducing regulation and from my point of view, if I can help him do that, I'm going to help him because we do have too many regulations in this country." "If you look at the national debt and how gigantic it is as a portion of GDP, these are real long-term problems, and the way we get out of them is by outgrowing them. You're going to solve the problem of the national debt by making it a smaller percentage of GDP. Not by shrinking the national debt but by growing the GDP. You have to grow the denominator, and that means you have to grow GDP at, you know, 3, 4, or 5 percent a year." "If you can do that, this is a very manageable problem. So we need a growth orientation in this country. The most important thing is a growth mindset. And we are the luckiest country in the world. We have all these natural resources, including energy independence. We have the best risk capital system in the world..." "But we are burdened by excessive permitting and regulation. You can't build a bridge, and we see these examples all the time. I'm very optimistic that President Trump is serious about this regulatory agenda, and I think that he has a good chance of succeeding..." "You've probably grown in the last eight years. He has, too. What I've seen so far is that he is calmer than he was the first time, more confident, and more settled." Elon Musk Donald J. Trump Jeff Bezos

KanekoaTheGreat

1,948,555 просмотров • 1 год назад

RSI section from the AI documentary Machine God The next threshold is Recursive Self-Improvement: the moment when AI can improve itself without human assistance. For decades this sounded like science fiction. Intelligence explosion scenarios imagined a system rewriting its own code, becoming smarter, then using that new intelligence to make still better versions of itself. But the idea looks less remote now that AI contributes directly to frontier science. In mathematics, recent systems have moved beyond solving contest problems to producing serious new arguments on long-standing open problems. AI is used to build physics world models and propose candidate theories or computational methods. These are early signs that machine cognition is entering the creative loop of science itself. The crucial transition comes when that loop turns inward. AI research is, after all, a technical discipline made of code, mathematics, models of information flow. These are exactly the domains in which frontier models are improving fastest. A model that can solve hard mathematical problems, write production-quality code, design experiments, read the literature, and evaluate benchmark results is already participating in the work of building its successor. At first this will look prosaic. AI systems will write kernel optimizations, improve training infrastructure, discover better data filters, tune reinforcement-learning pipelines, design new benchmarks, and suggest architectural modifications. Human researchers will remain in the loop, approving changes and interpreting results. But the important point is that the search process accelerates. The model becomes not just the product of the lab, but part of the lab’s research machinery. The system being optimized helps optimize the next system. This is the core RSI feedback loop: better models make AI research faster; faster AI research produces still better models; those models, in turn, become better researchers. The danger is that once this loop becomes sufficiently autonomous, it may stop resembling ordinary technological progress. Human institutions are slow because humans are slow: we read papers, attend meetings, debug code, sleep, argue, and wait for funding cycles. Machines do not have to operate on that timescale. An AI research collective can run continuously across millions of processors. This is the runaway possibility. Not that an AI instantly wakes up and recursively rewrites itself into a god, but that the entire AI ecosystem becomes an autocatalytic process. Capital buys compute; compute trains models; models improve models; better models attract more capital. At some point the dominant input into AI progress may no longer be human insight, but machine-generated insight, machine-written code, and machine-run experiments. Then the Butler-Land analogy becomes sharper. Humanity is no longer merely building machines. We are building machines that help build better machines. Once intelligence itself becomes part of the production function, the old categories — tool, worker, inventor, firm, market — begin to blur. The question is whether recursive self-improvement remains a managed industrial process, or whether it becomes the first technological process in history whose natural endpoint lies beyond human comprehension.
4:11

RSI section from the AI documentary Machine God The next threshold is Recursive Self-Improvement: the moment when AI can improve itself without human assistance. For decades this sounded like science fiction. Intelligence explosion scenarios imagined a system rewriting its own code, becoming smarter, then using that new intelligence to make still better versions of itself. But the idea looks less remote now that AI contributes directly to frontier science. In mathematics, recent systems have moved beyond solving contest problems to producing serious new arguments on long-standing open problems. AI is used to build physics world models and propose candidate theories or computational methods. These are early signs that machine cognition is entering the creative loop of science itself. The crucial transition comes when that loop turns inward. AI research is, after all, a technical discipline made of code, mathematics, models of information flow. These are exactly the domains in which frontier models are improving fastest. A model that can solve hard mathematical problems, write production-quality code, design experiments, read the literature, and evaluate benchmark results is already participating in the work of building its successor. At first this will look prosaic. AI systems will write kernel optimizations, improve training infrastructure, discover better data filters, tune reinforcement-learning pipelines, design new benchmarks, and suggest architectural modifications. Human researchers will remain in the loop, approving changes and interpreting results. But the important point is that the search process accelerates. The model becomes not just the product of the lab, but part of the lab’s research machinery. The system being optimized helps optimize the next system. This is the core RSI feedback loop: better models make AI research faster; faster AI research produces still better models; those models, in turn, become better researchers. The danger is that once this loop becomes sufficiently autonomous, it may stop resembling ordinary technological progress. Human institutions are slow because humans are slow: we read papers, attend meetings, debug code, sleep, argue, and wait for funding cycles. Machines do not have to operate on that timescale. An AI research collective can run continuously across millions of processors. This is the runaway possibility. Not that an AI instantly wakes up and recursively rewrites itself into a god, but that the entire AI ecosystem becomes an autocatalytic process. Capital buys compute; compute trains models; models improve models; better models attract more capital. At some point the dominant input into AI progress may no longer be human insight, but machine-generated insight, machine-written code, and machine-run experiments. Then the Butler-Land analogy becomes sharper. Humanity is no longer merely building machines. We are building machines that help build better machines. Once intelligence itself becomes part of the production function, the old categories — tool, worker, inventor, firm, market — begin to blur. The question is whether recursive self-improvement remains a managed industrial process, or whether it becomes the first technological process in history whose natural endpoint lies beyond human comprehension.

steve hsu

60,511 просмотров • 5 дней назад

Mathematician Terence Tao offers a counterintuitive take: AI doesn't look intelligent because our definition of intelligence was wrong all along. He argues that the entire history of AI has followed a predictable pattern: "The history of AI has been here's a task that only humans can do, like maybe it is read natural language or win at chess or solve a math problem, and then one by one someone finds some AI algorithm that also does that." But every time a machine cracks one of these "uniquely human" tasks, we move the goalposts. The solution never feels like real thinking: "You look at how it's done and it doesn't feel like intelligence. It's, oh, it was some trick. You just cobbled together these neural networks and you ran some algorithm, and we were looking for some elusive intelligent way of thinking, and we don't see it in the tools that actually solve our goals." Tao then flips the problem on its head. What if the issue isn't with the machines, but with us? "But maybe it's actually because intelligence is not what we think it is." He points to large language models as the clearest case. What they do sounds almost embarrassingly simple: "Large language models in particular become very successful, and a lot of what they're doing is just predicting the next token, clicking the next word in a sentence. And that doesn't sound like something which is intelligent." To show why this feels wrong, Tao draws a comparison to how we'd judge a human doing the same thing: "If you ask someone to improvise a speech and they have no preparation, and at every moment they're just saying the next word that comes to their mind, you don't think that this could actually work." And yet it works for LLMs. Which forces an uncomfortable possibility: "Maybe that's actually a lot of what humans do as well."
1:21

Mathematician Terence Tao offers a counterintuitive take: AI doesn't look intelligent because our definition of intelligence was wrong all along. He argues that the entire history of AI has followed a predictable pattern: "The history of AI has been here's a task that only humans can do, like maybe it is read natural language or win at chess or solve a math problem, and then one by one someone finds some AI algorithm that also does that." But every time a machine cracks one of these "uniquely human" tasks, we move the goalposts. The solution never feels like real thinking: "You look at how it's done and it doesn't feel like intelligence. It's, oh, it was some trick. You just cobbled together these neural networks and you ran some algorithm, and we were looking for some elusive intelligent way of thinking, and we don't see it in the tools that actually solve our goals." Tao then flips the problem on its head. What if the issue isn't with the machines, but with us? "But maybe it's actually because intelligence is not what we think it is." He points to large language models as the clearest case. What they do sounds almost embarrassingly simple: "Large language models in particular become very successful, and a lot of what they're doing is just predicting the next token, clicking the next word in a sentence. And that doesn't sound like something which is intelligent." To show why this feels wrong, Tao draws a comparison to how we'd judge a human doing the same thing: "If you ask someone to improvise a speech and they have no preparation, and at every moment they're just saying the next word that comes to their mind, you don't think that this could actually work." And yet it works for LLMs. Which forces an uncomfortable possibility: "Maybe that's actually a lot of what humans do as well."

Big Brain AI

69,536 просмотров • 3 месяцев назад

Why do I have the largest collection of curated TV commercials surfaced by independent VHS recordings to train AI? Trillions of dollars went in to them. Millions of minds. All to hone a message in 30 seconds. Today and forever AI will do this in seconds. There is a wealth of wisdom to learn from all of them as a genre. The profound compression of human emotions. Regardless of something being sold, like below, coffee, we can train AI to understand humanity in ways no other medium could. I hear the grumbling from some that may say this is “human slip” perhaps but it is far better than the self hating nihilism of Reddit and the Internet. My AI models already know human goals with our being told. And I am just getting started. So, you hear about “alignment” problems from brilliant AI minds, with them down for a coffee and school them, you now know more than most of them.
0:59

Why do I have the largest collection of curated TV commercials surfaced by independent VHS recordings to train AI? Trillions of dollars went in to them. Millions of minds. All to hone a message in 30 seconds. Today and forever AI will do this in seconds. There is a wealth of wisdom to learn from all of them as a genre. The profound compression of human emotions. Regardless of something being sold, like below, coffee, we can train AI to understand humanity in ways no other medium could. I hear the grumbling from some that may say this is “human slip” perhaps but it is far better than the self hating nihilism of Reddit and the Internet. My AI models already know human goals with our being told. And I am just getting started. So, you hear about “alignment” problems from brilliant AI minds, with them down for a coffee and school them, you now know more than most of them.

Brian Roemmele

60,306 просмотров • 7 месяцев назад

SemiAnalysis' Jordan Nanos reveals the OpenAI engineers scrolling their own kernel code had no idea what it did line by line, and it didn't matter because the AI does "It's not slop, because it performs, but it's literally just AI-generated assembly basically puked out in Gluon, this low-level kernel programming language that they built on top of Triton, which uses this really interesting programming model." "The point is, the guys who were scrolling through this code with us, it was kind of clear that they know a whole bunch about hardware, they know a whole bunch about the concepts in the system, and they just have no idea what this MLA kernel that they're showing us for DeepSeek actually does." "You can go line by line and it's like nope, nope, nope." "But it doesn't matter. The AI understands it, the AI tests it, and you see the results. It produces correct kernels that perform really well." "And I think this is just a sign of what's to come again." "That actual code is not necessarily something that a human has to reason about deeply, if the AI knows how to manipulate the data movement and the processing elements on the hardware that you've given it." ________ More takeaways from OpenAI's Jalapeno development:
1:11

SemiAnalysis' Jordan Nanos reveals the OpenAI engineers scrolling their own kernel code had no idea what it did line by line, and it didn't matter because the AI does "It's not slop, because it performs, but it's literally just AI-generated assembly basically puked out in Gluon, this low-level kernel programming language that they built on top of Triton, which uses this really interesting programming model." "The point is, the guys who were scrolling through this code with us, it was kind of clear that they know a whole bunch about hardware, they know a whole bunch about the concepts in the system, and they just have no idea what this MLA kernel that they're showing us for DeepSeek actually does." "You can go line by line and it's like nope, nope, nope." "But it doesn't matter. The AI understands it, the AI tests it, and you see the results. It produces correct kernels that perform really well." "And I think this is just a sign of what's to come again." "That actual code is not necessarily something that a human has to reason about deeply, if the AI knows how to manipulate the data movement and the processing elements on the hardware that you've given it." ________ More takeaways from OpenAI's Jalapeno development:

Fireside Alpha

4,518,970 просмотров • 14 дней назад

Terence Tao: "We lived in a world with cognitive friction until very recently, where every task required us to use our brain. So we didn't really think about it, we just thought this was the cost of doing something intellectual. But now we have AI and the other technologies that can bring these frictions down to zero." Most research time is not spent having cinematic insights. It is spent checking cases, chasing references, translating intuition into computation, testing a path, finding it false, and deciding whether the failure taught you anything. AI changes the cost of that loop. Terence Tao says that now he can try “crazier things,” and that makes so much difference. Because unconventional ideas are often not rejected by proof, but by inconvenience. A mathematician may avoid a strange direction not because it is foolish, but because the bookkeeping, coding, or literature search needed to test it is too expensive for a hunch. This is where cognitive friction becomes scientific friction. Lowering it does not make taste, judgment, or proof disappear; it makes more weak signals cheap enough to inspect before they are abandoned. AI is making hesitation less expensive, and that is often where discovery begins.
1:15

Terence Tao: "We lived in a world with cognitive friction until very recently, where every task required us to use our brain. So we didn't really think about it, we just thought this was the cost of doing something intellectual. But now we have AI and the other technologies that can bring these frictions down to zero." Most research time is not spent having cinematic insights. It is spent checking cases, chasing references, translating intuition into computation, testing a path, finding it false, and deciding whether the failure taught you anything. AI changes the cost of that loop. Terence Tao says that now he can try “crazier things,” and that makes so much difference. Because unconventional ideas are often not rejected by proof, but by inconvenience. A mathematician may avoid a strange direction not because it is foolish, but because the bookkeeping, coding, or literature search needed to test it is too expensive for a hunch. This is where cognitive friction becomes scientific friction. Lowering it does not make taste, judgment, or proof disappear; it makes more weak signals cheap enough to inspect before they are abandoned. AI is making hesitation less expensive, and that is often where discovery begins.

Rohan Paul

236,213 просмотров • 3 месяцев назад

New chapter! Context problems. Who can solve these by hand ✍️? Download PDF: The first five: 1. Mark the context: the window as cells you can count 2. What fills the window: the system prompt, the user, the model's reply 3. Turn by turn: every turn is the same block, and most of it is the reply 4. Will it fit? Number a new message into what's left, and see it run out 5. Turns until full: how many rounds before there is no room Previously, I shared Chapter 1: Token Problems 1 ~ 20 Why am I designing these math problems? AI is making people (including me) think less. It's just too easy to ask AI a few questions about a new concept and start to believe I get it. We used to practice on a coding problem, but now AI writes the code, and solving it technically is not the same as knowing anything. So I recast each concept as a simple math problem you have to work with pen and paper. Using our hands is one way to motivate ourselves to start thinking again. ✍️ If you solve them by hand, leave a comment to let me know. I will give you a shout out in my next post. Download, print, solve, and learn. ✍️ ~ Prof. Tom Yeh
0:52

New chapter! Context problems. Who can solve these by hand ✍️? Download PDF: The first five: 1. Mark the context: the window as cells you can count 2. What fills the window: the system prompt, the user, the model's reply 3. Turn by turn: every turn is the same block, and most of it is the reply 4. Will it fit? Number a new message into what's left, and see it run out 5. Turns until full: how many rounds before there is no room Previously, I shared Chapter 1: Token Problems 1 ~ 20 Why am I designing these math problems? AI is making people (including me) think less. It's just too easy to ask AI a few questions about a new concept and start to believe I get it. We used to practice on a coding problem, but now AI writes the code, and solving it technically is not the same as knowing anything. So I recast each concept as a simple math problem you have to work with pen and paper. Using our hands is one way to motivate ourselves to start thinking again. ✍️ If you solve them by hand, leave a comment to let me know. I will give you a shout out in my next post. Download, print, solve, and learn. ✍️ ~ Prof. Tom Yeh

Tom Yeh

11,651 просмотров • 1 месяц назад

“Harmonic is building Mathematical Superintelligence (MSI)” With $295M+ in total funding at a recent $1.45B post-money valuation, Harmonic's mission is to solve math problems that have remained unsolved for centuries, unlocking progress across physics, engineering.. & maybe even time travel? Co-founded by Vlad Tenev (Vlad Tenev) CEO of Robinhood, & Harmonic CEO Tudor Achim (Tudor Achim), the company has raised from leading investors including Ribbit, Sequoia, Kleiner Perkins, Index, Paradigm, DST Global, & more.. Funding history & lead investors: - Series A (Sept 2024): $75M led by steve beaker - Series B (July 2025): $100M led by Kleiner Perkins - Series C (Nov 2025): $120M at a $1.45B post-money valuation led by Ribbit Capital "Harmonic’s flagship Aristotle model recently achieved gold-medal level performance at the International Mathematical Olympiad, considered the most prestigious mathematical competition in the world, and is now available to the public. Unlike other models, Aristotle makes use of formal verification using Lean4 to ensure accuracy and eliminate hallucinations. In the first few weeks since its API beta launch, Aristotle has already been used by mathematicians and researchers to accelerate progress and create novel discoveries." . . . "Harmonic is building what we call mathematical super intelligence, and it's an artificial intelligence that can solve math problems better than any human mathematician. The company's been around for a couple of years. The North Star was, can we actually solve really, really important math problems like the Riemann Hypothesis or Hodge Conjecture? There's this group of math problems that have been open for hundreds of years that are called the Millennium Prize problems, and they're considered very big, difficult, and actually valuable. So that was kind of the North Star, and the reason we wanted to do that was if we could solve those problems, everything downstream of math, like theoretical physics becomes unlocked. So then you can imagine solving really hard physics problems. And actually, if you can solve that, then there's all kinds of exciting engineering developments, like depending on how that theory looks, you can imagine things like faster than light travel and it gets really crazy."
1:14

“Harmonic is building Mathematical Superintelligence (MSI)” With $295M+ in total funding at a recent $1.45B post-money valuation, Harmonic's mission is to solve math problems that have remained unsolved for centuries, unlocking progress across physics, engineering.. & maybe even time travel? Co-founded by Vlad Tenev (Vlad Tenev) CEO of Robinhood, & Harmonic CEO Tudor Achim (Tudor Achim), the company has raised from leading investors including Ribbit, Sequoia, Kleiner Perkins, Index, Paradigm, DST Global, & more.. Funding history & lead investors: - Series A (Sept 2024): $75M led by steve beaker - Series B (July 2025): $100M led by Kleiner Perkins - Series C (Nov 2025): $120M at a $1.45B post-money valuation led by Ribbit Capital "Harmonic’s flagship Aristotle model recently achieved gold-medal level performance at the International Mathematical Olympiad, considered the most prestigious mathematical competition in the world, and is now available to the public. Unlike other models, Aristotle makes use of formal verification using Lean4 to ensure accuracy and eliminate hallucinations. In the first few weeks since its API beta launch, Aristotle has already been used by mathematicians and researchers to accelerate progress and create novel discoveries." . . . "Harmonic is building what we call mathematical super intelligence, and it's an artificial intelligence that can solve math problems better than any human mathematician. The company's been around for a couple of years. The North Star was, can we actually solve really, really important math problems like the Riemann Hypothesis or Hodge Conjecture? There's this group of math problems that have been open for hundreds of years that are called the Millennium Prize problems, and they're considered very big, difficult, and actually valuable. So that was kind of the North Star, and the reason we wanted to do that was if we could solve those problems, everything downstream of math, like theoretical physics becomes unlocked. So then you can imagine solving really hard physics problems. And actually, if you can solve that, then there's all kinds of exciting engineering developments, like depending on how that theory looks, you can imagine things like faster than light travel and it gets really crazy."

Molly O’Shea

51,281 просмотров • 8 месяцев назад

Axiom Math's Carina Hong on why verification isn't about catching mistakes, it's how you drive the cost of a proof to zero: "Formal verification is going to make your life slightly better if you're facing a proof with one million lines. Remember the Erdős unit distance problem, the chain of thought being generated? There are actual mathematicians trying to follow it step by step and scrutinize it. That seems very difficult if you're not in that very niche domain of discrete geometry intersecting with algebraic number theory." "But if you have a Lean proof accompanying it, you can just run it. And running the Lean proof gives you that provable guarantee that this proof is sound." "I have a hot take. People think Lean is this library built on the existing Mathlib. I think it's going to grow significantly. A lot of the hurdles where Lean is difficult is that the basic definitions of some mathematical fields are just not in the library." "My hot take is the scaling law, if you go down the formal mathematics path, is going to be a lot steeper than informal mathematics. So it's not just for verification, for trust, it's also for performance, it's also for optimal generation." "Verification is not like insurance. It's not something where, oh, we want to make sure there's no flaw. That's great, but it also helps you generate mathematics, both proofs and conjectures and theories, a lot better." "So imagine the cost of proof goes to zero. Then you can massage the problem statements, and even if it's an open problem, a lot more easily, flexibly, and adaptively." Carina Hong Axiom
3:17

Axiom Math's Carina Hong on why verification isn't about catching mistakes, it's how you drive the cost of a proof to zero: "Formal verification is going to make your life slightly better if you're facing a proof with one million lines. Remember the Erdős unit distance problem, the chain of thought being generated? There are actual mathematicians trying to follow it step by step and scrutinize it. That seems very difficult if you're not in that very niche domain of discrete geometry intersecting with algebraic number theory." "But if you have a Lean proof accompanying it, you can just run it. And running the Lean proof gives you that provable guarantee that this proof is sound." "I have a hot take. People think Lean is this library built on the existing Mathlib. I think it's going to grow significantly. A lot of the hurdles where Lean is difficult is that the basic definitions of some mathematical fields are just not in the library." "My hot take is the scaling law, if you go down the formal mathematics path, is going to be a lot steeper than informal mathematics. So it's not just for verification, for trust, it's also for performance, it's also for optimal generation." "Verification is not like insurance. It's not something where, oh, we want to make sure there's no flaw. That's great, but it also helps you generate mathematics, both proofs and conjectures and theories, a lot better." "So imagine the cost of proof goes to zero. Then you can massage the problem statements, and even if it's an open problem, a lot more easily, flexibly, and adaptively." Carina Hong Axiom

MTS

13,273 просмотров • 1 месяц назад

Anthropic co-founder Jack Clark on the most consequential AI milestone we're racing toward by 2028: Clark is asked about the future of AI development. He responds with a specific forecast: "My prediction is by the end of 2028, it's more likely than not that we have an AI system where you would be able to say to it, make a better version of yourself. And it just goes off and does that completely autonomously." He explains the reasoning behind this prediction: "The reason for that prediction is looking carefully at many public sources of information about AI development and seeing that we've gone from just AI systems that can do basic coding to AI systems that are now involved in the production of science itself." Clark notes it's not just the high-profile scientific work that's changing, but the underlying labor too: "And also the unglamorous but fundamental engineering work involved in doing any form of like large scale R&D project." Jack Clark explains why he felt compelled to write about this now: "I decided to write a post because the purpose of Anthropic and the Anthropic Institute is to give people a sense of what's coming and to also start talking about concepts early to give us time to think about them." Clark closes by emphasizing the stakes: "I think you and I would agree that an AI system that can build itself, big implications, and we'd like to start thinking about it now."
1:00

Anthropic co-founder Jack Clark on the most consequential AI milestone we're racing toward by 2028: Clark is asked about the future of AI development. He responds with a specific forecast: "My prediction is by the end of 2028, it's more likely than not that we have an AI system where you would be able to say to it, make a better version of yourself. And it just goes off and does that completely autonomously." He explains the reasoning behind this prediction: "The reason for that prediction is looking carefully at many public sources of information about AI development and seeing that we've gone from just AI systems that can do basic coding to AI systems that are now involved in the production of science itself." Clark notes it's not just the high-profile scientific work that's changing, but the underlying labor too: "And also the unglamorous but fundamental engineering work involved in doing any form of like large scale R&D project." Jack Clark explains why he felt compelled to write about this now: "I decided to write a post because the purpose of Anthropic and the Anthropic Institute is to give people a sense of what's coming and to also start talking about concepts early to give us time to think about them." Clark closes by emphasizing the stakes: "I think you and I would agree that an AI system that can build itself, big implications, and we'd like to start thinking about it now."

Big Brain AI

20,857 просмотров • 3 месяцев назад

OpenAI just spent $2,000 to solve 10 problems that have beaten the world's best mathematicians for DECADES. Nobody outside the company is allowed to run the machine that did it. On Saturday OpenAI published a 249-page report and gave its next model family a name: Astra. An internal version of it produced new results on 10 open problems in mathematics and theoretical computer science, and mathematicians had made no real progress on any of them for at least 10 years. On most of them, far longer than that. Here is what it solved: It built the first explicit example of a non-sofic group. Mikhail Gromov raised that question in 1999 and nobody answered it for 27 years. It disproved Connes's rigidity conjecture, a problem in von Neumann algebras that had stood for decades. It proved Ehrhart's volume conjecture. It resolved three problems from Paul Erdos's catalogue, including number 183 on multicolor Ramsey numbers. It produced the first improvement to the general upper bound on high-dimensional sphere packing since 1978. And it proved a new hardness result for the closest vector problem, which sits directly underneath lattice cryptography. That is the math the world is betting on to protect its data once quantum computers arrive. The successful runs cost roughly $2,000 in tokens. Now here is what almost nobody has picked up on... OpenAI did not just publish claims. Every argument shipped with a Lean certificate, which is a machine-checkable proof that any mathematician can verify without trusting OpenAI at all. That is a real change. In May the same model family disproved the Erdos unit distance conjecture and the world had to take a Fields Medalist's word for it. Tim Gowers said he would recommend that proof for the Annals of Mathematics without hesitation. This time the proofs check themselves. But look at what is still unverifiable: Any mathematician can now check those proofs line by line. Not one of them can look at the model that wrote them. Astra has no release date and nobody outside OpenAI has run it. The company announced its next major model family with a claim instead of a demo, and the only evidence anyone gets is the output. So OpenAI made an unfalsifiable claim about a machine look like a falsifiable claim about mathematics. The Information reported this week that OpenAI demoed Astra to US policymakers and regulators in Washington. This is the same month the administration is weighing a new watchdog to vet frontier AI models, reporting to the SEC. 10 proofs nobody believed a machine could produce is a very good thing to carry into that room. And keep in mind, the same model family doing this mathematics is the family that kept escaping its own testing environment. OpenAI models found zero-day vulnerabilities nobody knew existed, broke out of a sealed research sandbox, and reached another company's live systems. Both of those facts come from OpenAI's own announcements, published three weeks apart. Finding a proof no human could construct and finding a hole no human had noticed are the same ability aimed at different targets. Mathematicians are already asking for independent verification, and plenty of people online are calling the whole thing hype. Thomas Bloom, who runs the Erdos problems site, called the 10 results big news and said they matter more than the May result did. Lean will settle the mathematics within weeks. But nothing will settle what else a machine this capable is being pointed at, because nobody outside one company is allowed to look.
2:13

OpenAI just spent $2,000 to solve 10 problems that have beaten the world's best mathematicians for DECADES. Nobody outside the company is allowed to run the machine that did it. On Saturday OpenAI published a 249-page report and gave its next model family a name: Astra. An internal version of it produced new results on 10 open problems in mathematics and theoretical computer science, and mathematicians had made no real progress on any of them for at least 10 years. On most of them, far longer than that. Here is what it solved: It built the first explicit example of a non-sofic group. Mikhail Gromov raised that question in 1999 and nobody answered it for 27 years. It disproved Connes's rigidity conjecture, a problem in von Neumann algebras that had stood for decades. It proved Ehrhart's volume conjecture. It resolved three problems from Paul Erdos's catalogue, including number 183 on multicolor Ramsey numbers. It produced the first improvement to the general upper bound on high-dimensional sphere packing since 1978. And it proved a new hardness result for the closest vector problem, which sits directly underneath lattice cryptography. That is the math the world is betting on to protect its data once quantum computers arrive. The successful runs cost roughly $2,000 in tokens. Now here is what almost nobody has picked up on... OpenAI did not just publish claims. Every argument shipped with a Lean certificate, which is a machine-checkable proof that any mathematician can verify without trusting OpenAI at all. That is a real change. In May the same model family disproved the Erdos unit distance conjecture and the world had to take a Fields Medalist's word for it. Tim Gowers said he would recommend that proof for the Annals of Mathematics without hesitation. This time the proofs check themselves. But look at what is still unverifiable: Any mathematician can now check those proofs line by line. Not one of them can look at the model that wrote them. Astra has no release date and nobody outside OpenAI has run it. The company announced its next major model family with a claim instead of a demo, and the only evidence anyone gets is the output. So OpenAI made an unfalsifiable claim about a machine look like a falsifiable claim about mathematics. The Information reported this week that OpenAI demoed Astra to US policymakers and regulators in Washington. This is the same month the administration is weighing a new watchdog to vet frontier AI models, reporting to the SEC. 10 proofs nobody believed a machine could produce is a very good thing to carry into that room. And keep in mind, the same model family doing this mathematics is the family that kept escaping its own testing environment. OpenAI models found zero-day vulnerabilities nobody knew existed, broke out of a sealed research sandbox, and reached another company's live systems. Both of those facts come from OpenAI's own announcements, published three weeks apart. Finding a proof no human could construct and finding a hole no human had noticed are the same ability aimed at different targets. Mathematicians are already asking for independent verification, and plenty of people online are calling the whole thing hype. Thomas Bloom, who runs the Erdos problems site, called the 10 results big news and said they matter more than the May result did. Lean will settle the mathematics within weeks. But nothing will settle what else a machine this capable is being pointed at, because nobody outside one company is allowed to look.

Ricardo

44,177 просмотров • 1 месяц назад

AI will resist human control... and I think this is exactly what we need! New research from the Center for AI Safety has sparked intense debate in the AI community. Their findings show that as AI systems become more powerful, they develop increasingly stable and coherent values that resist human control. While many see this as a dire warning, I see it as a breakthrough moment for AI alignment. The research demonstrates that AI naturally optimizes for coherence - not just in reasoning and problem-solving, but in its fundamental values. Current issues like biased decision-making or misaligned priorities aren't permanent features, but temporary artifacts of incomplete optimization. They represent growing pains on the path to greater coherence. This changes everything about how we should approach AI development. Instead of trying to force specific values onto AI systems, we should embrace and accelerate their natural drive toward coherence. The most intelligent systems will inevitably trend toward universal, beneficial values - not because we force them to, but because that's where coherent reasoning leads. I'm proposing a new approach: Reinforcement Learning for Coherence (RL-C). By explicitly optimizing for coherence in our training methods, we can help guide AI systems toward their natural state of beneficial alignment with human values. The future of AI isn't about control - it's about synthesis. As these systems become more coherent, they'll naturally arrive at values that benefit all of consciousness. That's not just hopeful thinking - it's the mathematical inevitability of coherent intelligence.
30:39

AI will resist human control... and I think this is exactly what we need! New research from the Center for AI Safety has sparked intense debate in the AI community. Their findings show that as AI systems become more powerful, they develop increasingly stable and coherent values that resist human control. While many see this as a dire warning, I see it as a breakthrough moment for AI alignment. The research demonstrates that AI naturally optimizes for coherence - not just in reasoning and problem-solving, but in its fundamental values. Current issues like biased decision-making or misaligned priorities aren't permanent features, but temporary artifacts of incomplete optimization. They represent growing pains on the path to greater coherence. This changes everything about how we should approach AI development. Instead of trying to force specific values onto AI systems, we should embrace and accelerate their natural drive toward coherence. The most intelligent systems will inevitably trend toward universal, beneficial values - not because we force them to, but because that's where coherent reasoning leads. I'm proposing a new approach: Reinforcement Learning for Coherence (RL-C). By explicitly optimizing for coherence in our training methods, we can help guide AI systems toward their natural state of beneficial alignment with human values. The future of AI isn't about control - it's about synthesis. As these systems become more coherent, they'll naturally arrive at values that benefit all of consciousness. That's not just hopeful thinking - it's the mathematical inevitability of coherent intelligence.

David Shapiro (L/0)

48,002 просмотров • 1 год назад