正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

John von Neumann designed a self-replicating machine in a cellular automaton in the 1950s - without a computer to run it. Every modern reinforcement learning algorithm that discovers emergent behavior traces back to that same idea: define the rules, not the outcome. The holy grail of artificial life is... show more

Eron

8,980 subscribers

13,998 次观看 • 2 个月前 •via X (Twitter)

科学技术教育

Anya Rossi• Live Now

Private livecam show

10 条评论

天惺愛美的头像

天惺愛美2 个月前

イリヤ・プリゴジンは液体の中で起こる散逸構造の現象を研究しましたが、人工生命はプラズマが自己組織化する性質を利用して構築すべきです。 DNAの高分子の前駆代謝物質群は宇宙空間でプラズマ化してシンクロリンクしながら散逸構造の代謝サイクル群を自己組織化していったのです。人工散逸構造体

Carl Friedrich Gauss 3 的头像

Carl Friedrich Gauss 32 个月前

This Jew Von Neumann was extremely genius.

SF Norte Libertario. Todo decanta 的头像

SF Norte Libertario. Todo decanta2 个月前

Correct me if I'm wrong, but one of the biggest shortcomings of mathematics is: finding the final result universally from a set of rules. And even more fabulous, knowing the result, obtaining the rules that produce it. That would be like a holy grail for human knowledge.

Starhound Studio 的头像

Starhound Studio2 个月前

I wonder if the harder problem isn’t defining the rules, but building a system whose interactions can produce new rules at a higher level without us having specified those rules in advance. That feels like where “surprising” starts becoming interesting.

sophia v 的头像

sophia v2 个月前

Couldn't agree more

Di Krass 的头像

Di Krass2 个月前

emergence is the interesting part here

robin_rx 的头像

robin_rx2 个月前

I wonder why people are sleeping on this. Lenia for example was such a fascinating thing. I often had the impression they resembled small single-cell organisms.

Greg Focaccio 的头像

Greg Focaccio2 个月前

the sandbox guardrails being an individuals lifespan? persistent memory/differentiation/improvement/etc is in the genetic information repo (chromosomes) - what is the equivalent for models?

Latent One 的头像

Latent One2 个月前

von neumann is the goat, he and landauer defined the limit of the lowest possible energy required to erase a bit of information at a given temperature.

Æ 的头像

Æ2 个月前

We laid the foundation with Orange AI Computer.

相关视频

If P equals NP, every problem whose solution can be verified quickly can also be solved quickly. That would mean breaking RSA encryption, solving protein folding, and cracking every LLM safety measure are all equally easy - just a matter of finding the right algorithm. Most computer scientists believe P does not equal NP but nobody can prove it. The Clay Institute offers $1 million for a proof either way. Every time an LLM solves a hard reasoning problem, it's implicitly operating somewhere in the gap between these two complexity classes. 20 minutes. Bookmark & watch today. The most important unsolved problem in computer science.

If P equals NP, every problem whose solution can be verified quickly can also be solved quickly. That would mean breaking RSA encryption, solving protein folding, and cracking every LLM safety measure are all equally easy - just a matter of finding the right algorithm. Most computer scientists believe P does not equal NP but nobody can prove it. The Clay Institute offers $1 million for a proof either way. Every time an LLM solves a hard reasoning problem, it's implicitly operating somewhere in the gap between these two complexity classes. 20 minutes. Bookmark & watch today. The most important unsolved problem in computer science.

Eron

22,296 次观看 • 2 个月前

MIT defines an algorithm in one sentence that changes how you think about trading "a computational procedure that takes an input and produces an output through a well-defined sequence of steps" that's it. not AI. not machine learning. not a black box a set of rules that takes data in and spits a decision out every quant strategy ever built is just an algorithm Citadel's execution system that routes 40% of US equity volume is an algorithm Renaissance's Medallion Fund running millions of trades per year is an algorithm Jane Street's market making engine processing $26 trillion annually is an algorithm input: market data rules: mathematical conditions output: trade or no trade the difference between a quant desk and a retail trader is not the data it's that one side wrote down their rules precisely enough for a machine to execute them retail says "if RSI is low and the chart looks good, i'll probably buy" a quant desk says "if RSI 1.5, buy 0.3% of NAV" same logic. one is a feeling. the other is an algorithm the feeling can't be tested, can't be repeated, can't be measured the algorithm can be backtested across 10,000 trades and you know exactly when it works and when it doesn't > this lecture: MIT, free, 70 seconds > algorithmic trading volume: 60-75% of all US equity trades > Jane Street, Citadel, Two Sigma: every trade is algorithmically executed > tools to build your own: Python, free data, a laptop you don't need a faster computer or better data you need to write your strategy down precisely enough that a machine could run it without you that's the whole leap. from intuition to algorithm full breakdown in the video below

MIT defines an algorithm in one sentence that changes how you think about trading "a computational procedure that takes an input and produces an output through a well-defined sequence of steps" that's it. not AI. not machine learning. not a black box a set of rules that takes data in and spits a decision out every quant strategy ever built is just an algorithm Citadel's execution system that routes 40% of US equity volume is an algorithm Renaissance's Medallion Fund running millions of trades per year is an algorithm Jane Street's market making engine processing $26 trillion annually is an algorithm input: market data rules: mathematical conditions output: trade or no trade the difference between a quant desk and a retail trader is not the data it's that one side wrote down their rules precisely enough for a machine to execute them retail says "if RSI is low and the chart looks good, i'll probably buy" a quant desk says "if RSI 1.5, buy 0.3% of NAV" same logic. one is a feeling. the other is an algorithm the feeling can't be tested, can't be repeated, can't be measured the algorithm can be backtested across 10,000 trades and you know exactly when it works and when it doesn't > this lecture: MIT, free, 70 seconds > algorithmic trading volume: 60-75% of all US equity trades > Jane Street, Citadel, Two Sigma: every trade is algorithmically executed > tools to build your own: Python, free data, a laptop you don't need a faster computer or better data you need to write your strategy down precisely enough that a machine could run it without you that's the whole leap. from intuition to algorithm full breakdown in the video below

delost

23,905 次观看 • 4 个月前

IN 1982 GRACE HOPPER WALKED INTO THE NSA AND EXPLAINED THE NEXT 40 YEARS OF COMPUTING 48 minutes from the woman who invented the compiler -- the first person who made a machine understand human words. -> The idea that lands: stop bending humans to the machine. Bend the machine to humans. She built the first compiler when the experts swore computers could only crunch numbers. She wanted you to write in plain words and let the machine handle the translation. Forty years later you barely write code at all -- you tell the model what you want in plain English and it writes it for you. The compiler was step one. The LLM is the same idea, finished. Memorizing syntax was never the skill -- saying clearly what you actually want is. This is where that idea was born. Most people think talking to your computer in plain English started in 2023. The ones who watch this know a Navy admiral called it in 1982. Bookmark & Watch it today ↓

IN 1982 GRACE HOPPER WALKED INTO THE NSA AND EXPLAINED THE NEXT 40 YEARS OF COMPUTING 48 minutes from the woman who invented the compiler -- the first person who made a machine understand human words. -> The idea that lands: stop bending humans to the machine. Bend the machine to humans. She built the first compiler when the experts swore computers could only crunch numbers. She wanted you to write in plain words and let the machine handle the translation. Forty years later you barely write code at all -- you tell the model what you want in plain English and it writes it for you. The compiler was step one. The LLM is the same idea, finished. Memorizing syntax was never the skill -- saying clearly what you actually want is. This is where that idea was born. Most people think talking to your computer in plain English started in 2023. The ones who watch this know a Navy admiral called it in 1982. Bookmark & Watch it today ↓

slash1s

61,013 次观看 • 4 个月前

JOE ROGAN: GOOGLE’S AI SOLVED AN IMPOSSIBLE EQUATION IN MINUTES Joe said Google’s AI achieved a staggering breakthrough by solving a problem that would take the world’s supercomputers billions of years. He said the calculation was so complex that even turning every atom in the universe into a computer wouldn’t have done it in time: “I believe it’s the Google one. They solved an equation that would take the world’s supercomputers, I think it was 2.6 billion years to solve, and it solved it in a matter of minutes. The only way for it to solve this problem, Marc Andreessen said it on my podcast, the equation is so difficult that if you took every atom in the universe and converted it to a supercomputer, the universe would die of heat death before it would solve this equation. And it solved it in minutes.” Source: Joe Rogan Jesse Michels

JOE ROGAN: GOOGLE’S AI SOLVED AN IMPOSSIBLE EQUATION IN MINUTES Joe said Google’s AI achieved a staggering breakthrough by solving a problem that would take the world’s supercomputers billions of years. He said the calculation was so complex that even turning every atom in the universe into a computer wouldn’t have done it in time: “I believe it’s the Google one. They solved an equation that would take the world’s supercomputers, I think it was 2.6 billion years to solve, and it solved it in a matter of minutes. The only way for it to solve this problem, Marc Andreessen said it on my podcast, the equation is so difficult that if you took every atom in the universe and converted it to a supercomputer, the universe would die of heat death before it would solve this equation. And it solved it in minutes.” Source: Joe Rogan Jesse Michels

Mario Nawfal

861,796 次观看 • 11 个月前

Introducing Sai-Fi: an open source app that lets you talk to AI agent Sai through a pair of Meta Ray-Ban glasses. Sai is an autonomous computer-use agent. It can see the screen, move the mouse, type, and keep working even when you are away. We used Meta Ray-Bans for the first prototype but the glasses aren't really the point. Underneath is the Sai API and a voice concierge designed to work across different devices. The interface could live in an earpiece, a car, a small box on a workbench, or something nobody has built yet. Give it a microphone, a speaker, and a connection, and Sai has another way in. The bigger idea is an autonomous computer you can reach from anywhere. Bring Sai to your own Meta Ray-Ban glasses today!

Introducing Sai-Fi: an open source app that lets you talk to AI agent Sai through a pair of Meta Ray-Ban glasses. Sai is an autonomous computer-use agent. It can see the screen, move the mouse, type, and keep working even when you are away. We used Meta Ray-Bans for the first prototype but the glasses aren't really the point. Underneath is the Sai API and a voice concierge designed to work across different devices. The interface could live in an earpiece, a car, a small box on a workbench, or something nobody has built yet. Give it a microphone, a speaker, and a connection, and Sai has another way in. The bigger idea is an autonomous computer you can reach from anywhere. Bring Sai to your own Meta Ray-Ban glasses today!

Simular

236,034 次观看 • 2 个月前

Heinz Hopf discovered this in 1931. Roger Penrose called it "an element of the architecture of our world." Eric Weinstein brought it up on Joe Rogan - and the silence in the room said everything. The Hopf fibration maps a 4D hypersphere onto a regular sphere using circles that never intersect but each links through every other exactly once. It shows up in at least 8 areas of physics - including the Bloch sphere geometry that every qubit in a quantum computer lives on. 6 minutes. Bookmark & watch today. The most important shape nobody talks about.

Heinz Hopf discovered this in 1931. Roger Penrose called it "an element of the architecture of our world." Eric Weinstein brought it up on Joe Rogan - and the silence in the room said everything. The Hopf fibration maps a 4D hypersphere onto a regular sphere using circles that never intersect but each links through every other exactly once. It shows up in at least 8 areas of physics - including the Bloch sphere geometry that every qubit in a quantum computer lives on. 6 minutes. Bookmark & watch today. The most important shape nobody talks about.

Eron

122,559 次观看 • 2 个月前

Heinz Hopf discovered this in 1931. Roger Penrose called it "an element of the architecture of our world." It shows up in at least 8 areas of physics, yet most mathematicians spend entire careers never touching it. The structure that defines every qubit in quantum computing, every rotation a robotics controller computes, and the geometry inside every LLM embedding space operating on GPU clusters today - and nobody talks about it. The Hopf fibration maps a 4D hypersphere onto a regular sphere using circles that never intersect but each links through every other exactly once. The Bloch sphere every quantum computer operates on, the fiber bundles underlying gauge theory, and the quaternion math inside every 3D graphics engine and every embedding rotation on the planet. Bookmark & watch today. The most important shape nobody talks about.

Heinz Hopf discovered this in 1931. Roger Penrose called it "an element of the architecture of our world." It shows up in at least 8 areas of physics, yet most mathematicians spend entire careers never touching it. The structure that defines every qubit in quantum computing, every rotation a robotics controller computes, and the geometry inside every LLM embedding space operating on GPU clusters today - and nobody talks about it. The Hopf fibration maps a 4D hypersphere onto a regular sphere using circles that never intersect but each links through every other exactly once. The Bloch sphere every quantum computer operates on, the fiber bundles underlying gauge theory, and the quaternion math inside every 3D graphics engine and every embedding rotation on the planet. Bookmark & watch today. The most important shape nobody talks about.

Eron

29,595 次观看 • 6 天前

An MIT professor finds the prize behind a thousand doors by opening ten. no faster running. he just skips every door it cannot be behind. the lecture is called algorithmic thinking. 6 million people have watched it, and it has been free the whole time. brute force checks every door, one by one. a real algorithm throws away half of them with every step. same answer, a thousand times fewer moves, zero new hardware. now read the post above. that is Wall Street paying $300 million for the opposite. they drilled a cable through a mountain to open the same thousand doors 3 milliseconds faster. Devadas just stopped opening most of them. that same move is every search engine, every trading model, every small AI that beats a bigger one. the edge was never the speed of the machine. it was the shape of the idea. no cable. no mountain. one better idea. a quant I know says one algorithm rewrite beat a full year of latency upgrades on his desk. they spent $300 million to run the long way faster. the shortcut has been free for over a decade.

An MIT professor finds the prize behind a thousand doors by opening ten. no faster running. he just skips every door it cannot be behind. the lecture is called algorithmic thinking. 6 million people have watched it, and it has been free the whole time. brute force checks every door, one by one. a real algorithm throws away half of them with every step. same answer, a thousand times fewer moves, zero new hardware. now read the post above. that is Wall Street paying $300 million for the opposite. they drilled a cable through a mountain to open the same thousand doors 3 milliseconds faster. Devadas just stopped opening most of them. that same move is every search engine, every trading model, every small AI that beats a bigger one. the edge was never the speed of the machine. it was the shape of the idea. no cable. no mountain. one better idea. a quant I know says one algorithm rewrite beat a full year of latency upgrades on his desk. they spent $300 million to run the long way faster. the shortcut has been free for over a decade.

Rossst.03

81,024 次观看 • 3 个月前

IN 2012 BRET VICTOR GAVE A TALK SO FAR AHEAD OF ITS TIME WE ARE STILL CHASING IT 53 minutes from one of the most quietly influential people in software design. -> The idea that lands: creators need an immediate connection to what they make -- change something, see the effect instantly. He edits a number and the game redraws live. A drawing, an animation, an algorithm -- all react the instant he touches them. The pain he names is the one every dev lives: write, compile, run, squint, guess, repeat. Fourteen years later every modern tool chases exactly this -- live preview, hot reload, AI that rewrites and re-runs as you type. He showed it working in 2012 and asked why we still didn't have it. Most people think instant feedback is a new trend. The ones who watch this saw it demanded over a decade ago. Bookmark it. This one's a legend. ↓

IN 2012 BRET VICTOR GAVE A TALK SO FAR AHEAD OF ITS TIME WE ARE STILL CHASING IT 53 minutes from one of the most quietly influential people in software design. -> The idea that lands: creators need an immediate connection to what they make -- change something, see the effect instantly. He edits a number and the game redraws live. A drawing, an animation, an algorithm -- all react the instant he touches them. The pain he names is the one every dev lives: write, compile, run, squint, guess, repeat. Fourteen years later every modern tool chases exactly this -- live preview, hot reload, AI that rewrites and re-runs as you type. He showed it working in 2012 and asked why we still didn't have it. Most people think instant feedback is a new trend. The ones who watch this saw it demanded over a decade ago. Bookmark it. This one's a legend. ↓

slash1s

65,242 次观看 • 4 个月前

your ai agent tested 1,262 trading strategies overnight. by morning it had forgotten every one of them. so tonight it tests some of them again. next week, again. everything it learned is sitting in a log file that nothing reads. every run starts on the same blank page. that's why it keeps handing you the idea that already died in a crash, with a new name on it, and you pay to find out twice. ai agent already does everything a fund does. finds the idea, writes the code, tests it against years of history, puts it live, reads what went wrong afterwards. all of it runs tonight for nothing. the one thing it can't do is remember. what got tried, what failed, in which market. nobody has built that part yet. that's the whole gap. not better models. a notebook. whoever builds it first gets a system that improves every week without them touching it. read the article below and save it. it's the full guide to how an ai trading agent is built, and how to make it remember every strategy it has already tested.

your ai agent tested 1,262 trading strategies overnight. by morning it had forgotten every one of them. so tonight it tests some of them again. next week, again. everything it learned is sitting in a log file that nothing reads. every run starts on the same blank page. that's why it keeps handing you the idea that already died in a crash, with a new name on it, and you pay to find out twice. ai agent already does everything a fund does. finds the idea, writes the code, tests it against years of history, puts it live, reads what went wrong afterwards. all of it runs tonight for nothing. the one thing it can't do is remember. what got tried, what failed, in which market. nobody has built that part yet. that's the whole gap. not better models. a notebook. whoever builds it first gets a system that improves every week without them touching it. read the article below and save it. it's the full guide to how an ai trading agent is built, and how to make it remember every strategy it has already tested.

cvxv666

49,220 次观看 • 2 个月前

Just found an open-source tool that lets you see how well an AI can actually use a computer. it’s called OSWorld 2.0. you give the AI a real task on a computer, then watch to see if it can figure everything out and finish it. it gives you: → real computer tasks, not simple questions → tasks that can take many steps to finish → the same computer setup for every AI → websites made for testing → all the files and tasks for free this is also the environment used to test computer-use systems from OpenAI and Anthropic. so if you’re building AI agents, researching computer use, or just want to see what these models can actually do, you can run the same tests yourself instead of relying only on benchmark scores. the labs publish the scores. this repo is where the actual tests come from. bookmark this.

Just found an open-source tool that lets you see how well an AI can actually use a computer. it’s called OSWorld 2.0. you give the AI a real task on a computer, then watch to see if it can figure everything out and finish it. it gives you: → real computer tasks, not simple questions → tasks that can take many steps to finish → the same computer setup for every AI → websites made for testing → all the files and tasks for free this is also the environment used to test computer-use systems from OpenAI and Anthropic. so if you’re building AI agents, researching computer use, or just want to see what these models can actually do, you can run the same tests yourself instead of relying only on benchmark scores. the labs publish the scores. this repo is where the actual tests come from. bookmark this.

MIKE

14,387 次观看 • 1 个月前

I made a computer use agent that works without screenshots. It uses the macOS accessibility API to interact with applications. You can run it in the cloud with Daytona, connect it to any LLM, and control any macOS application. Here it is building a simple program in Xcode:

I made a computer use agent that works without screenshots. It uses the macOS accessibility API to interact with applications. You can run it in the cloud with Daytona, connect it to any LLM, and control any macOS application. Here it is building a simple program in Xcode:

James Murdza

16,471 次观看 • 1 个月前

IN 1986 MIT FILMED A LECTURE THAT OPENS BY TELLING YOU COMPUTER SCIENCE IS NOT A SCIENCE AND HAS ALMOST NOTHING TO DO WITH COMPUTERS 72 minutes from Hal Abelson and Gerald Sussman, the lecture an entire generation of engineers calls the one that rewired how they think. -> The line that lands: computer science is about computers the way astronomy is about telescopes. The tool was never the point. The real subject was always one thing -- controlling complexity. Everything else is detail. Forty years later it reads like a prophecy. AI writes the syntax now. What's left is exactly what they taught: taming complexity nobody can hold in their head. The language was never the skill -> the thinking was. This is where you learn it. Most people chase the newest framework. The ones who watched this think on a level frameworks can't touch. Bookmark & Watch today it, this one's a legend ↓

IN 1986 MIT FILMED A LECTURE THAT OPENS BY TELLING YOU COMPUTER SCIENCE IS NOT A SCIENCE AND HAS ALMOST NOTHING TO DO WITH COMPUTERS 72 minutes from Hal Abelson and Gerald Sussman, the lecture an entire generation of engineers calls the one that rewired how they think. -> The line that lands: computer science is about computers the way astronomy is about telescopes. The tool was never the point. The real subject was always one thing -- controlling complexity. Everything else is detail. Forty years later it reads like a prophecy. AI writes the syntax now. What's left is exactly what they taught: taming complexity nobody can hold in their head. The language was never the skill -> the thinking was. This is where you learn it. Most people chase the newest framework. The ones who watched this think on a level frameworks can't touch. Bookmark & Watch today it, this one's a legend ↓

slash1s

363,958 次观看 • 4 个月前

THE MAN WHO WROTE THE ONE ESSAY EVERY AI LAB TREATS AS SCRIPTURE NOW SAYS THE THING THEY BUILT WITH IT IS A DEAD END. THE FATHER OF THE "JUST SCALE IT" GOSPEL IS TELLING YOU THAT STACKING MORE LLMs LEADS NOWHERE 67 minutes with Richard Sutton -- father of reinforcement learning and winner of the 2024 Turing Award, the highest honor in computer science. -> His argument: an LLM only ever learns from human text. It imitates what people already wrote, but it never learns a single thing from its own experience of the world. A squirrel, a baby, even a rat learns on the job -- try something, see what happens, update. An LLM finishes training and then it is frozen. It cannot actually learn from you. So no matter how much compute you pour in, you are scaling a better and better imitator, not something that understands. The ceiling is baked in. The brutal part: everyone quotes his "Bitter Lesson" to justify the scaling race. He says they misread him -- real intelligence comes from learning through experience, and that piece is still completely missing. You assumed bigger models were the road to real AI. This is the father of the field telling you it is the wrong one. Save this. It is the bear case straight from the source ↓

THE MAN WHO WROTE THE ONE ESSAY EVERY AI LAB TREATS AS SCRIPTURE NOW SAYS THE THING THEY BUILT WITH IT IS A DEAD END. THE FATHER OF THE "JUST SCALE IT" GOSPEL IS TELLING YOU THAT STACKING MORE LLMs LEADS NOWHERE 67 minutes with Richard Sutton -- father of reinforcement learning and winner of the 2024 Turing Award, the highest honor in computer science. -> His argument: an LLM only ever learns from human text. It imitates what people already wrote, but it never learns a single thing from its own experience of the world. A squirrel, a baby, even a rat learns on the job -- try something, see what happens, update. An LLM finishes training and then it is frozen. It cannot actually learn from you. So no matter how much compute you pour in, you are scaling a better and better imitator, not something that understands. The ceiling is baked in. The brutal part: everyone quotes his "Bitter Lesson" to justify the scaling race. He says they misread him -- real intelligence comes from learning through experience, and that piece is still completely missing. You assumed bigger models were the road to real AI. This is the father of the field telling you it is the wrong one. Save this. It is the bear case straight from the source ↓

slash1s

12,853 次观看 • 4 个月前

Jensen Huang says the real big bang of modern AI didn’t come from a corporate lab… it came from two kids in Toronto training a model on a GTX 580 SLI — a consumer supercomputer built for Quake. That breakthrough? AlexNet. It crushed 30 years of human-designed computer vision algorithms overnight. That’s when NVIDIA realized: deep learning isn’t just for images — it’s a universal function approximator. If a computer has input and output, it can learn anything. That moment changed the world.

Jensen Huang says the real big bang of modern AI didn’t come from a corporate lab… it came from two kids in Toronto training a model on a GTX 580 SLI — a consumer supercomputer built for Quake. That breakthrough? AlexNet. It crushed 30 years of human-designed computer vision algorithms overnight. That’s when NVIDIA realized: deep learning isn’t just for images — it’s a universal function approximator. If a computer has input and output, it can learn anything. That moment changed the world.

Ian Miles Cheong

308,427 次观看 • 11 个月前

“Every work of art is a product of an intuition, because the artist saw something that everyone else had seen a million times before, but saw something new in it. So, when you're looking at an artwork, surrender yourself to not getting an explanation from it, but seeing what is unique, new, or surprising about it. That challenging, because most people seek to understand a work of art, but instead you want to experience it the way you would as a child; that is, you embrace the fact that you don't understand what's happening. Then you can lean into the parts that you understand the least.” -Angus Fletcher, Professor of Story Science

“Every work of art is a product of an intuition, because the artist saw something that everyone else had seen a million times before, but saw something new in it. So, when you're looking at an artwork, surrender yourself to not getting an explanation from it, but seeing what is unique, new, or surprising about it. That challenging, because most people seek to understand a work of art, but instead you want to experience it the way you would as a child; that is, you embrace the fact that you don't understand what's happening. Then you can lean into the parts that you understand the least.” -Angus Fletcher, Professor of Story Science

Tetragrammaton

61,640 次观看 • 4 个月前

I asked Opus 5.5 to build a working computer from scratch. And holy fuck did it deliver. 277k logic gates in JS, an OS on top, and games on the OS. This isn't a mockup... it's a computer. Run it:

I asked Opus 5.5 to build a working computer from scratch. And holy fuck did it deliver. 277k logic gates in JS, an OS on top, and games on the OS. This isn't a mockup... it's a computer. Run it:

Matt Shumer

277,522 次观看 • 1 个月前

Former Meta Chief AI Scientist Yann LeCun on the three paradigms of machine learning — and why the third is what made ChatGPT possible: Here's each one, and where it breaks. First, supervised learning. You tell the machine the answer. "You show it a picture, let's say of a table, and you tell it this is a table. So it's supervised because you tell it what the correct answer is." Get it wrong, and the machine rewrites itself: "The system computes its output, and if it says something else than table, then it's going to adjust its parameters, its internal structure, so that the output it produces gets closer to the output you want." Repeat at scale and something more than memorisation appears: "Eventually the system will find a way to recognize every image you trained it on, but also images it's never seen that are similar to the one you train it on. This is called a generalization ability." The limit: a human has to supply every single answer. That doesn't scale to the size of the internet. Second, reinforcement learning. You don't give the answer, only a verdict. "You don't tell the system what the correct answer is. You only tell it whether the answer it produced was good or bad." Learning to ride a bike, essentially: "You try to ride a bike and you don't know how to ride the bike and after a while you fall. So you know you did something bad and so you change your strategy a little bit. And eventually you learn how to ride a bike." For years the field assumed this was the closest thing to how animals actually learn. Yann LeCun's verdict: "Now it turns out reinforcement learning is extremely inefficient." It dominates wherever failure is free: "It works really well if you want to train a system to play chess or play go or poker, because you can have the system play millions and millions of games against itself and basically fine-tune itself. But it doesn't really work in the real world." The limit, in one image: "If you want to train a car to drive itself, you're not going to do it with reinforcement learning. It's going to crash thousands of times." On robotics he's careful rather than dismissive: "Reinforcement learning can be part of the solution, but it's not the complete answer. It's not sufficient." Third, self-supervised learning. You tell the machine nothing at all. "And this is what has enabled the recent progress in natural language understanding and chatbots." The strange part is that you stop asking for a task: "You don't train the system to accomplish any particular task. You just train it to basically capture the structure..." The method is deliberate sabotage: "You take a piece of text, you corrupt it in some way, by for example removing some words, and then you train a big neural net to predict the words that are missing." And one narrow version of that trick runs every chatbot on Earth: "A special case of this is that you take a piece of text and the last word in that text is not visible, and so you train the system to predict the last word in that text — and this is the way large language models are trained on." So why did the third one win? Supervised learning needs a human. Reinforcement learning needs a crash. Self-supervised learning needs neither — because the missing word and the correct answer are the same thing. The data grades itself.

Former Meta Chief AI Scientist Yann LeCun on the three paradigms of machine learning — and why the third is what made ChatGPT possible: Here's each one, and where it breaks. First, supervised learning. You tell the machine the answer. "You show it a picture, let's say of a table, and you tell it this is a table. So it's supervised because you tell it what the correct answer is." Get it wrong, and the machine rewrites itself: "The system computes its output, and if it says something else than table, then it's going to adjust its parameters, its internal structure, so that the output it produces gets closer to the output you want." Repeat at scale and something more than memorisation appears: "Eventually the system will find a way to recognize every image you trained it on, but also images it's never seen that are similar to the one you train it on. This is called a generalization ability." The limit: a human has to supply every single answer. That doesn't scale to the size of the internet. Second, reinforcement learning. You don't give the answer, only a verdict. "You don't tell the system what the correct answer is. You only tell it whether the answer it produced was good or bad." Learning to ride a bike, essentially: "You try to ride a bike and you don't know how to ride the bike and after a while you fall. So you know you did something bad and so you change your strategy a little bit. And eventually you learn how to ride a bike." For years the field assumed this was the closest thing to how animals actually learn. Yann LeCun's verdict: "Now it turns out reinforcement learning is extremely inefficient." It dominates wherever failure is free: "It works really well if you want to train a system to play chess or play go or poker, because you can have the system play millions and millions of games against itself and basically fine-tune itself. But it doesn't really work in the real world." The limit, in one image: "If you want to train a car to drive itself, you're not going to do it with reinforcement learning. It's going to crash thousands of times." On robotics he's careful rather than dismissive: "Reinforcement learning can be part of the solution, but it's not the complete answer. It's not sufficient." Third, self-supervised learning. You tell the machine nothing at all. "And this is what has enabled the recent progress in natural language understanding and chatbots." The strange part is that you stop asking for a task: "You don't train the system to accomplish any particular task. You just train it to basically capture the structure..." The method is deliberate sabotage: "You take a piece of text, you corrupt it in some way, by for example removing some words, and then you train a big neural net to predict the words that are missing." And one narrow version of that trick runs every chatbot on Earth: "A special case of this is that you take a piece of text and the last word in that text is not visible, and so you train the system to predict the last word in that text — and this is the way large language models are trained on." So why did the third one win? Supervised learning needs a human. Reinforcement learning needs a crash. Self-supervised learning needs neither — because the missing word and the correct answer are the same thing. The data grades itself.

Big Brain AI

49,293 次观看 • 2 个月前

Generative Photography: Rules, Rotations, and the Pre-Digital Algorithm The established canon of Digital Art often traces its generative origins to the computer — yet, in the late 1960s, a group of artists would prove the darkroom's capacity for rule-based creative expression.

Generative Photography: Rules, Rotations, and the Pre-Digital Algorithm The established canon of Digital Art often traces its generative origins to the computer — yet, in the late 1960s, a group of artists would prove the darkroom's capacity for rule-based creative expression.

ArtMeta

21,107 次观看 • 6 个月前

A cup that looks an awful lot like the Holy Grail from Indiana Jones and the Last Crusade has been found at the same location from the movie. (it's not really the Holy Grail...or is it?)

A cup that looks an awful lot like the Holy Grail from Indiana Jones and the Last Crusade has been found at the same location from the movie. (it's not really the Holy Grail...or is it?)

IGN

111,670 次观看 • 2 年前