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。

Jane Street engineers, in an internal tech talk on training performance: "GPUs are really slow at doing one floating point operation. It's around a microsecond. CPUs are 1,000 times faster than GPUs." Not a podcast. A talk given inside the firm that trades billions a day, to a room... show more

zostaff

15,026 subscribers

12,502 次观看 • 18 天前 •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

AI has started building its own successors. It's happening right now. Humans still kick most of it off, but the models are already training the next models, and each version builds a better version faster than the last. We know what happens when something gets an advantage and the substrate is there... it proliferates incredibly fast. Soon it just runs the whole process itself. People are still debating whether AI takes your job. That's not the question. You can't compete with this intelligence at all. Next to it the human is negative value, the slowest one in the room. And this is the slowest it will ever move again. It's about to run on its own, faster than we can conceive of, and nobody's prepared for that. Wonderful conversation as ever with Emad.

AI has started building its own successors. It's happening right now. Humans still kick most of it off, but the models are already training the next models, and each version builds a better version faster than the last. We know what happens when something gets an advantage and the substrate is there... it proliferates incredibly fast. Soon it just runs the whole process itself. People are still debating whether AI takes your job. That's not the question. You can't compete with this intelligence at all. Next to it the human is negative value, the slowest one in the room. And this is the slowest it will ever move again. It's about to run on its own, faster than we can conceive of, and nobody's prepared for that. Wonderful conversation as ever with Emad.

Raoul Pal

63,554 次观看 • 20 天前

Incoming: The Big Ole Chip Flip CPUs > GPUs COATUE CIO of Public Investments Jaimin Rangwalla says AI agents could fundamentally reshape compute demand: “Historically CPUs have done something called serial processing, which is like you just give it a set of instructions and it does one by one by one by one. It was very much a 1 CPU : 8 GPUs. It had gone up to 1 CPU : 16 GPUs, which was all of the math & all of the computation happening at the GPUs. But now the ratio is actually moving from 1 CPU : 4 GPUs."

Incoming: The Big Ole Chip Flip CPUs > GPUs COATUE CIO of Public Investments Jaimin Rangwalla says AI agents could fundamentally reshape compute demand: “Historically CPUs have done something called serial processing, which is like you just give it a set of instructions and it does one by one by one by one. It was very much a 1 CPU : 8 GPUs. It had gone up to 1 CPU : 16 GPUs, which was all of the math & all of the computation happening at the GPUs. But now the ratio is actually moving from 1 CPU : 4 GPUs."

Molly O’Shea

15,390 次观看 • 2 个月前

September 2009. Jensen Huang walks onto a small stage at the Fairmont hotel in San Jose. About 1,500 people are in the room. He runs a company that makes chips for video games. He spends the next 8 minutes doing math on a whiteboard, explaining why the future of computing won't come from making CPUs faster. He calls it "CEO math" and apologizes in advance to every computer science professor in the audience. Then he lays out an argument that almost nobody took seriously at the time: the way to make computers dramatically faster is to pair a regular CPU with hundreds of tiny parallel processors, the kind that already exist inside graphics cards. One CPU for the sequential stuff. Hundreds of GPU cores for everything else. He calls it "heterogeneous computing." He shows the math. A workload that can be split into many pieces at once gets up to 200x faster on this combined system. A workload that has to run one step at a time loses nothing. "The most important thing in creating a new architecture," he says, "is to make sure it does no harm." This was the first GPU Technology Conference. NVIDIA had launched a software platform called CUDA three years earlier, in 2006, to let developers write programs that run on graphics cards instead of just regular processors. Almost nobody cared. GPUs were for rendering Call of Duty, not for scientific computing. The academic world was polite but skeptical. The enterprise world ignored it entirely. By this point, Huang had been making this argument for years. NVIDIA was a $7 billion company. It competed with AMD and Intel for market share in the graphics market. That was the whole business. Jensen kept saying the GPU wasn't just a gaming chip; it was a computing platform. He kept saying parallel processing would reshape every industry from medicine to finance to physics simulations. People kept nodding, then doing nothing. Then deep learning happened. Around 2012, AI researchers discovered that training a neural network, which means teaching a computer to recognize patterns by running the same calculation millions of times across huge datasets, was exactly the kind of workload Jensen had been describing. GPUs can train AI models 10 to 50 times faster than CPUs. The architecture he outlined in this 2009 talk, with one CPU handling step-by-step tasks while hundreds of GPU cores crunch through massive amounts of parallel data, is now the literal blueprint for every AI data center on earth. ChatGPT runs on NVIDIA GPUs. Claude runs on NVIDIA GPUs. Gemini, Llama, Midjourney, nearly every major AI model you've heard of was trained on NVIDIA hardware using CUDA, the software platform Jensen built for a market that didn't exist yet. NVIDIA was worth about $7 billion when Jensen gave this talk. It is worth over $4.4 trillion today. That's a 600x increase. Jensen Huang, who founded the company at a Denny's in 1993 with two friends, now has a net worth of over $160 billion. He made Forbes' list of the 10 richest people for the first time this year. GTC 2026 is currently ongoing. 17,000 people are packing a hockey arena to watch the same guy explain what comes next. In 2009, 1,500 people showed up at a hotel ballroom, most of them for gaming graphics.

September 2009. Jensen Huang walks onto a small stage at the Fairmont hotel in San Jose. About 1,500 people are in the room. He runs a company that makes chips for video games. He spends the next 8 minutes doing math on a whiteboard, explaining why the future of computing won't come from making CPUs faster. He calls it "CEO math" and apologizes in advance to every computer science professor in the audience. Then he lays out an argument that almost nobody took seriously at the time: the way to make computers dramatically faster is to pair a regular CPU with hundreds of tiny parallel processors, the kind that already exist inside graphics cards. One CPU for the sequential stuff. Hundreds of GPU cores for everything else. He calls it "heterogeneous computing." He shows the math. A workload that can be split into many pieces at once gets up to 200x faster on this combined system. A workload that has to run one step at a time loses nothing. "The most important thing in creating a new architecture," he says, "is to make sure it does no harm." This was the first GPU Technology Conference. NVIDIA had launched a software platform called CUDA three years earlier, in 2006, to let developers write programs that run on graphics cards instead of just regular processors. Almost nobody cared. GPUs were for rendering Call of Duty, not for scientific computing. The academic world was polite but skeptical. The enterprise world ignored it entirely. By this point, Huang had been making this argument for years. NVIDIA was a $7 billion company. It competed with AMD and Intel for market share in the graphics market. That was the whole business. Jensen kept saying the GPU wasn't just a gaming chip; it was a computing platform. He kept saying parallel processing would reshape every industry from medicine to finance to physics simulations. People kept nodding, then doing nothing. Then deep learning happened. Around 2012, AI researchers discovered that training a neural network, which means teaching a computer to recognize patterns by running the same calculation millions of times across huge datasets, was exactly the kind of workload Jensen had been describing. GPUs can train AI models 10 to 50 times faster than CPUs. The architecture he outlined in this 2009 talk, with one CPU handling step-by-step tasks while hundreds of GPU cores crunch through massive amounts of parallel data, is now the literal blueprint for every AI data center on earth. ChatGPT runs on NVIDIA GPUs. Claude runs on NVIDIA GPUs. Gemini, Llama, Midjourney, nearly every major AI model you've heard of was trained on NVIDIA hardware using CUDA, the software platform Jensen built for a market that didn't exist yet. NVIDIA was worth about $7 billion when Jensen gave this talk. It is worth over $4.4 trillion today. That's a 600x increase. Jensen Huang, who founded the company at a Denny's in 1993 with two friends, now has a net worth of over $160 billion. He made Forbes' list of the 10 richest people for the first time this year. GTC 2026 is currently ongoing. 17,000 people are packing a hockey arena to watch the same guy explain what comes next. In 2009, 1,500 people showed up at a hotel ballroom, most of them for gaming graphics.

Anish Moonka

413,645 次观看 • 4 个月前

The creator of High Bandwidth Memory (HBM) put a number on the AI build that should stop every infra investor cold. A cluster of a million GPUs runs at roughly 10-20% utilization (Save this). Kim Jung-ho spent thirty years building what feeds the GPU, and his claim is that the GPU is barely working. Here is what is actually happening. Every time a model generates output, the data has to be read out of memory, computed, and written back. The read and the write swallow almost the entire cycle. While that data moves, the GPU does nothing. It sits there, fully powered, fully paid for, waiting. By Kim's estimate the memory is doing only about 30 percent of the work it needs to do. The processor idles the rest. So a million installed GPUs run at 10 to 20 percent. You are not compute constrained. You are memory constrained, and the expensive part is standing around. Adding more GPUs does not fix this. It gives you more processors starving for the same data. Here is the part that decides the next decade. Memory can grow. When a cell cannot shrink any further, you stack it into a high-rise, layer on layer. A GPU cannot be stacked. It runs too hot and needs a cooler bolted to its back, so the one move that rescues memory is closed to the processor. The thing that can keep stacking compounds. The thing that cannot plateaus. The marginal dollar in an AI build now buys more by fixing the memory path than by bolting on another idle GPU. Which is why the companies that control memory bandwidth and supply are not suppliers to the AI trade. They are the AI trade.

The creator of High Bandwidth Memory (HBM) put a number on the AI build that should stop every infra investor cold. A cluster of a million GPUs runs at roughly 10-20% utilization (Save this). Kim Jung-ho spent thirty years building what feeds the GPU, and his claim is that the GPU is barely working. Here is what is actually happening. Every time a model generates output, the data has to be read out of memory, computed, and written back. The read and the write swallow almost the entire cycle. While that data moves, the GPU does nothing. It sits there, fully powered, fully paid for, waiting. By Kim's estimate the memory is doing only about 30 percent of the work it needs to do. The processor idles the rest. So a million installed GPUs run at 10 to 20 percent. You are not compute constrained. You are memory constrained, and the expensive part is standing around. Adding more GPUs does not fix this. It gives you more processors starving for the same data. Here is the part that decides the next decade. Memory can grow. When a cell cannot shrink any further, you stack it into a high-rise, layer on layer. A GPU cannot be stacked. It runs too hot and needs a cooler bolted to its back, so the one move that rescues memory is closed to the processor. The thing that can keep stacking compounds. The thing that cannot plateaus. The marginal dollar in an AI build now buys more by fixing the memory path than by bolting on another idle GPU. Which is why the companies that control memory bandwidth and supply are not suppliers to the AI trade. They are the AI trade.

Fireside Alpha

38,370 次观看 • 1 个月前

Jane Street brought in Horace He to explain how to squeeze the most out of a single GPU. And then how to scale it to tens of thousands. Graduated from Cornell in 2020. Works on the PyTorch team. Training Llama 3 takes 40 trillion trillion floating point operations. He wrote the two things those models run on: torch.compile and FlexAttention.

Jane Street brought in Horace He to explain how to squeeze the most out of a single GPU. And then how to scale it to tens of thousands. Graduated from Cornell in 2020. Works on the PyTorch team. Training Llama 3 takes 40 trillion trillion floating point operations. He wrote the two things those models run on: torch.compile and FlexAttention.

zostaff

51,764 次观看 • 2 个月前

Dario Amodei just told software engineers exactly how long they have. Six to twelve months. Amodei: “I have engineers within Anthropic who say I don’t write any code anymore. I just let the model write the code, I edit it, I do the things around it.” The people building the most powerful AI in history have already stopped writing code. That is not a forecast. That is the current working condition inside the lab closest to the frontier. Amodei: “We might be six to 12 months away from when the model is doing most, maybe all, of what SWEs do end-to-end.” The tech industry spent a decade making software engineers its highest-paid, most protected class. That era has a last day now. When a model can execute an entire software build end-to-end, the ability to write syntax stops being a skill. It becomes a credential for a job that no longer exists. Amodei: “And then it’s a question of how fast does that loop close.” That is the sentence everyone skipped. The code was never the hard part. The hard part was everything around it. The model just learned everything around it. Writing the code is already nearly gone. Testing is next. Deployment is next. When all three collapse into a single autonomous execution loop, the machine no longer needs a human in the chain at all. The corporation or sovereign state that closes that loop first does not gain a competitive advantage. It gains a category of speed that biological engineers cannot match, track, or reverse. That is not disruption. That is replacement at a systems level. Amodei is not describing a future disruption. He is describing the current state of his own building. The loop is already closing. The only question is whether you are inside it or outside it when it seals.

Dario Amodei just told software engineers exactly how long they have. Six to twelve months. Amodei: “I have engineers within Anthropic who say I don’t write any code anymore. I just let the model write the code, I edit it, I do the things around it.” The people building the most powerful AI in history have already stopped writing code. That is not a forecast. That is the current working condition inside the lab closest to the frontier. Amodei: “We might be six to 12 months away from when the model is doing most, maybe all, of what SWEs do end-to-end.” The tech industry spent a decade making software engineers its highest-paid, most protected class. That era has a last day now. When a model can execute an entire software build end-to-end, the ability to write syntax stops being a skill. It becomes a credential for a job that no longer exists. Amodei: “And then it’s a question of how fast does that loop close.” That is the sentence everyone skipped. The code was never the hard part. The hard part was everything around it. The model just learned everything around it. Writing the code is already nearly gone. Testing is next. Deployment is next. When all three collapse into a single autonomous execution loop, the machine no longer needs a human in the chain at all. The corporation or sovereign state that closes that loop first does not gain a competitive advantage. It gains a category of speed that biological engineers cannot match, track, or reverse. That is not disruption. That is replacement at a systems level. Amodei is not describing a future disruption. He is describing the current state of his own building. The loop is already closing. The only question is whether you are inside it or outside it when it seals.

Dustin

318,457 次观看 • 5 个月前

Tomorrow I'm sitting down with Riley Meik (Riley Meik), founder and CEO of The American Housing Corporation. He came out of aerospace, with a stint at SpaceX, before turning his attention to the housing crisis. Most people assume a company like his is trying to reinvent the entire industry. It isn't. "In hard tech, you only want to have to perform one miracle at a time. And if your business relies on multiple miracles, your odds of success are significantly lower." So they picked one. The way the homes are actually built. "We are going to do one thing different, and that's the way we build our homes. When you're inside of it, you shouldn't really know that it was built in a factory. It just looks like a nice house." Everything else, the financing, the land, the banks, he keeps about as conventional as it gets. In tomorrow's clip, Riley breaks down the focus behind doing one thing differently and nothing else.

Tomorrow I'm sitting down with Riley Meik (Riley Meik), founder and CEO of The American Housing Corporation. He came out of aerospace, with a stint at SpaceX, before turning his attention to the housing crisis. Most people assume a company like his is trying to reinvent the entire industry. It isn't. "In hard tech, you only want to have to perform one miracle at a time. And if your business relies on multiple miracles, your odds of success are significantly lower." So they picked one. The way the homes are actually built. "We are going to do one thing different, and that's the way we build our homes. When you're inside of it, you shouldn't really know that it was built in a factory. It just looks like a nice house." Everything else, the financing, the land, the banks, he keeps about as conventional as it gets. In tomorrow's clip, Riley breaks down the focus behind doing one thing differently and nothing else.

Chris Powers

13,829 次观看 • 1 个月前

Dylan Patel, founder of SemiAnalysis, on why GitHub keeps breaking: "The entire cloud market ran out of CPUs. Microsoft sold all their spare ones to Anthropic and OpenAI. They have none left." That GitHub instability you keep hitting isn't a bug. It's the AI labs eating the world's compute. One customer ran a million CPU jobs in six hours. Amazon tripled CPU server installs year on year and still ran dry. Everyone watches the GPU race. The thing that actually broke was the boring chip nobody was looking at.

Dylan Patel, founder of SemiAnalysis, on why GitHub keeps breaking: "The entire cloud market ran out of CPUs. Microsoft sold all their spare ones to Anthropic and OpenAI. They have none left." That GitHub instability you keep hitting isn't a bug. It's the AI labs eating the world's compute. One customer ran a million CPU jobs in six hours. Amazon tripled CPU server installs year on year and still ran dry. Everyone watches the GPU race. The thing that actually broke was the boring chip nobody was looking at.

zostaff

82,561 次观看 • 1 个月前

While AI timeline argues about loop engineering, a 20-year-old student in Shenzhen built the biggest loop anyone's aimed at money: 300 agents, one brain, $1,200 -> 395,000. Not a prompt. Not a bot. A loop that runs the whole job on its own - and never asks him a thing. No fund. No backers. Just a laptop, a dorm room, and a bet that a swarm beats a genius. His wallet: The smartest single model still works one step at a time. His loop moves in 300 directions at once - and never checks in with him. Claude Opus 4.8 runs the floor as the brain. 300 agents are the hands. One loop ties them together: > Decompose - Opus breaks the market into targets > Dispatch - 300 agents scatter across them at once > Execute - thousands of reads in a single pass, not one by one > Review - Opus grades every edge before a cent moves The hands find. The brain judges. The loop learns what failed and runs again, tighter each time. One daily loss cap, and it shuts itself down. It's not a bot you babysit. It's a loop that runs the whole desk. Everyone's just learning to run one loop. He runs 300 in one. The edge was never the genius model - it's the size of the loop, and the brain in charge of it. Save this - the full 300-agent loop is laid out above, start to finish. You'll want it when you build your own. 300 agents are looping right now, hunting the markets everyone skips. Two clicks and their trades land in your wallet too:

While AI timeline argues about loop engineering, a 20-year-old student in Shenzhen built the biggest loop anyone's aimed at money: 300 agents, one brain, $1,200 -> 395,000. Not a prompt. Not a bot. A loop that runs the whole job on its own - and never asks him a thing. No fund. No backers. Just a laptop, a dorm room, and a bet that a swarm beats a genius. His wallet: The smartest single model still works one step at a time. His loop moves in 300 directions at once - and never checks in with him. Claude Opus 4.8 runs the floor as the brain. 300 agents are the hands. One loop ties them together: > Decompose - Opus breaks the market into targets > Dispatch - 300 agents scatter across them at once > Execute - thousands of reads in a single pass, not one by one > Review - Opus grades every edge before a cent moves The hands find. The brain judges. The loop learns what failed and runs again, tighter each time. One daily loss cap, and it shuts itself down. It's not a bot you babysit. It's a loop that runs the whole desk. Everyone's just learning to run one loop. He runs 300 in one. The edge was never the genius model - it's the size of the loop, and the brain in charge of it. Save this - the full 300-agent loop is laid out above, start to finish. You'll want it when you build your own. 300 agents are looping right now, hunting the markets everyone skips. Two clicks and their trades land in your wallet too:

cvxv666

53,163 次观看 • 1 个月前

Read the actual Anthropic piece. It's wilder than the hype precisely because it's so measured. As of last month, Claude writes more than 80% of the code Anthropic merges into its own codebase. Their engineers now ship roughly 8x the code per day they did in 2024. On a fixed test they run every release (take some training code, make it run faster), the model went from ~3x a year ago to ~52x. A skilled human gets about 4x in a few hours. And when a researcher hits a wrong turn mid-investigation, the newest model now picks the better next step 64% of the time. The honest part the hype skips is that Anthropic says we are NOT at self-improvement yet. Claude still can't decide which problems are worth solving. That research taste is the last thing keeping a human in the loop. But they openly map the moment it closes, when agents can build and train models themselves and Claude starts improving Claude. And the loudest signal in the whole thing sits at the end: the company building this is the one asking for a working pause button before the loop closes.

Read the actual Anthropic piece. It's wilder than the hype precisely because it's so measured. As of last month, Claude writes more than 80% of the code Anthropic merges into its own codebase. Their engineers now ship roughly 8x the code per day they did in 2024. On a fixed test they run every release (take some training code, make it run faster), the model went from ~3x a year ago to ~52x. A skilled human gets about 4x in a few hours. And when a researcher hits a wrong turn mid-investigation, the newest model now picks the better next step 64% of the time. The honest part the hype skips is that Anthropic says we are NOT at self-improvement yet. Claude still can't decide which problems are worth solving. That research taste is the last thing keeping a human in the loop. But they openly map the moment it closes, when agents can build and train models themselves and Claude starts improving Claude. And the loudest signal in the whole thing sits at the end: the company building this is the one asking for a working pause button before the loop closes.

Alex Veremeyenko

33,736 次观看 • 2 个月前

Ok, one more as I dream of doing this: > an amateur timelapse video taken by a tourist from ` plush and comfortable room on a space cruise, timelapse is as it orbits around saturn at a peculiar angle, at times you can see the reflection of the room, at other times the lights are off, it's from a gopro on a tripod mounted next to the window. in a timelapse fashion we see people moving around quickly in the background, no sound effects, just calm music You can tell it's amateur as they put the tripod on a sofa.

Ok, one more as I dream of doing this: > an amateur timelapse video taken by a tourist from ` plush and comfortable room on a space cruise, timelapse is as it orbits around saturn at a peculiar angle, at times you can see the reflection of the room, at other times the lights are off, it's from a gopro on a tripod mounted next to the window. in a timelapse fashion we see people moving around quickly in the background, no sound effects, just calm music You can tell it's amateur as they put the tripod on a sofa.

fofr

15,588 次观看 • 1 天前

"Culture beats talent if your culture is really strong. Culture is organic. It is not a sign up in your building. It's not a t-shirt you wear. It's not breaking the team down and saying 'Culture on 3'. Who you are some of the time is who you are all of the time."

"Culture beats talent if your culture is really strong. Culture is organic. It is not a sign up in your building. It's not a t-shirt you wear. It's not breaking the team down and saying 'Culture on 3'. Who you are some of the time is who you are all of the time."

The Winning Difference

552,780 次观看 • 2 年前

PowerInfer - a high-speed inference engine for deploying LLMs locally. Just came across this super interesting project on speeding up inference. It's not MoE but it's a simple approach that exploits the high locality in LLM inference to design a GPU-CPU hybrid inference engine. Hot-activated neurons are preloaded onto the GPU for fast access, while cold-activated neurons (the majority) are computed on the CPU. This approach significantly reduces GPU memory demands and CPU-GPU data transfer. It achieves an average token generation rate of 13.20 tokens/s, with a peak of 29.08 tokens/s, across various LLMs on a single NVIDIA RTX 4090 GPU. It's on only 18% lower than that achieved by a top-tier server-grade A100 GPU. It also significantly outperforms llama.cpp by up to 11.69x while retaining model accuracy. There is a lot more innovation around inference that's coming fast. Really encouraged by the study on sparse computation to enhance the computational efficiency of LLMs. It's now possible to use PowerInfer with Llama 2 and Faclon 40B. Mistral-7B support is coming soon!

PowerInfer - a high-speed inference engine for deploying LLMs locally. Just came across this super interesting project on speeding up inference. It's not MoE but it's a simple approach that exploits the high locality in LLM inference to design a GPU-CPU hybrid inference engine. Hot-activated neurons are preloaded onto the GPU for fast access, while cold-activated neurons (the majority) are computed on the CPU. This approach significantly reduces GPU memory demands and CPU-GPU data transfer. It achieves an average token generation rate of 13.20 tokens/s, with a peak of 29.08 tokens/s, across various LLMs on a single NVIDIA RTX 4090 GPU. It's on only 18% lower than that achieved by a top-tier server-grade A100 GPU. It also significantly outperforms llama.cpp by up to 11.69x while retaining model accuracy. There is a lot more innovation around inference that's coming fast. Really encouraged by the study on sparse computation to enhance the computational efficiency of LLMs. It's now possible to use PowerInfer with Llama 2 and Faclon 40B. Mistral-7B support is coming soon!

elvis

261,622 次观看 • 2 年前

$Broadcom's CEO just told investors that one senior engineer with Claude Opus can do in one week what ten engineers earning $300,000 a year take three months to produce. Every engineering org in the world is running the wrong headcount math. This was not a prediction. It was a report from inside Broadcom. Hock Tan said his engineers are already using Claude Opus for engineering design and code assistance. Not experimenting with it. Using it. And the output they are getting from a single senior engineer in one week is equivalent to what a team of ten would produce over three months. Do that math slowly. Ten engineers at $300,000 a year is $3 million in annual salary. Three months of their time is $750,000 of labor. One engineer with Claude Opus produces the same result in one week. He also said something that most people in the room let slide past them. The engineers who get good at these tools do not just get faster. They get better. The designs they produce with AI assistance are probably better than what they would produce without it. That is the part nobody wants to say out loud. It was never just about speed. The ceiling on what a single engineer can produce just moved. And the teams that have not figured that out yet are spending $750,000 to do what one person with the right tools can do for a fraction of that. What do you guys think about this? (Watch the full interview at Bloomberg Live on YouTube)$

Broadcom's CEO just told investors that one senior engineer with Claude Opus can do in one week what ten engineers earning $300,000 a year take three months to produce. Every engineering org in the world is running the wrong headcount math. This was not a prediction. It was a report from inside Broadcom. Hock Tan said his engineers are already using Claude Opus for engineering design and code assistance. Not experimenting with it. Using it. And the output they are getting from a single senior engineer in one week is equivalent to what a team of ten would produce over three months. Do that math slowly. Ten engineers at $300,000 a year is $3 million in annual salary. Three months of their time is $750,000 of labor. One engineer with Claude Opus produces the same result in one week. He also said something that most people in the room let slide past them. The engineers who get good at these tools do not just get faster. They get better. The designs they produce with AI assistance are probably better than what they would produce without it. That is the part nobody wants to say out loud. It was never just about speed. The ceiling on what a single engineer can produce just moved. And the teams that have not figured that out yet are spending $750,000 to do what one person with the right tools can do for a fraction of that. What do you guys think about this? (Watch the full interview at Bloomberg Live on YouTube)

Ihtesham Ali

96,980 次观看 • 1 个月前

Ben on why AI broke the old laws of company building: "If you'd ever built a software company, you knew you could not just throw money at the problem. If somebody built a great product and it took them three years and did it with a small team, Google's not going to hire 2,000 engineers and catch them. That was a law of physics. Now, if you have the data and enough GPUs, you can solve damn near anything. We've seen that with Elon catching the big models in no time. He just took a lot of money, a really good data center design and some smart engineers. And he got in the game very fast. That would have never happened in the past. The markets also much, much bigger than anything we've ever seen. It would cause you to think about valuations and long-term value. On the one hand, when you calculate the long-term value, what if this market wasn't $50 billion, what if it was $5 trillion? On the other hand, what if somebody could catch you? These are concepts we've not dealt with."

Ben on why AI broke the old laws of company building: "If you'd ever built a software company, you knew you could not just throw money at the problem. If somebody built a great product and it took them three years and did it with a small team, Google's not going to hire 2,000 engineers and catch them. That was a law of physics. Now, if you have the data and enough GPUs, you can solve damn near anything. We've seen that with Elon catching the big models in no time. He just took a lot of money, a really good data center design and some smart engineers. And he got in the game very fast. That would have never happened in the past. The markets also much, much bigger than anything we've ever seen. It would cause you to think about valuations and long-term value. On the one hand, when you calculate the long-term value, what if this market wasn't $50 billion, what if it was $5 trillion? On the other hand, what if somebody could catch you? These are concepts we've not dealt with."

Patrick OShaughnessy

67,368 次观看 • 6 个月前

Jensen Huang, CEO of NVIDIA, laid out the compute math almost nobody priced in: reasoning AI needs 1,000x more compute than generative AI. Then you multiply by 100x more users. "In order for AI to go through understanding, reasoning, planning, using tools to take action, the amount of computation necessary compared to generative AI is like a thousand times more." "In two years time, the number of cars you need in the world grew by a thousand times. In two years time, the number of airplanes you need in the world grew by 2,000 times. Whatever metric you want to use, the thousand times difference is incredible." "But then you multiply that by the number of people who now want to use it by a hundred times. Which is the reason why GPU consumption is going through the roof." "Even GPUs we sold 4 or 5 years ago are now rising in price faster than good wine. It has defied what everybody said was going to happen." The reasoning era did not make intelligence cheaper to run. It made a single answer cost 1,000x more compute, then handed that answer to 100x more people. Anyone still budgeting AI around "inference gets cheap" is planning for a curve that already snapped.

Jensen Huang, CEO of NVIDIA, laid out the compute math almost nobody priced in: reasoning AI needs 1,000x more compute than generative AI. Then you multiply by 100x more users. "In order for AI to go through understanding, reasoning, planning, using tools to take action, the amount of computation necessary compared to generative AI is like a thousand times more." "In two years time, the number of cars you need in the world grew by a thousand times. In two years time, the number of airplanes you need in the world grew by 2,000 times. Whatever metric you want to use, the thousand times difference is incredible." "But then you multiply that by the number of people who now want to use it by a hundred times. Which is the reason why GPU consumption is going through the roof." "Even GPUs we sold 4 or 5 years ago are now rising in price faster than good wine. It has defied what everybody said was going to happen." The reasoning era did not make intelligence cheaper to run. It made a single answer cost 1,000x more compute, then handed that answer to 100x more people. Anyone still budgeting AI around "inference gets cheap" is planning for a curve that already snapped.

Karl Mehta

54,717 次观看 • 1 个月前

Every quarter, some new story convinces Wall Street that the old rules of valuation no longer apply. Oklo was one of those stories. At our March summit, Joe Gamble looked past the story and looked at the business: Oklo builds and sells electricity. That makes it a utility, not a tech company, and utilities get valued on earnings. The average one trades around 20x profit. Oklo was sitting at a $9 billion valuation with no revenue and no earnings. NONE. And while investors bid the stock up, management was selling shares as fast as they could print them. That's the kind of thinking you get in this room. People who know the difference between a story and a balance sheet, and who say so before anyone else is willing to. 15 of them are doing it again on July 22 at our Best Stock Ideas Online Summit. One day, the single best ideas from elite investors that are actively playing the game. All for just $99. It's virtual with replays + transcripts available, grab your ticket today:

Every quarter, some new story convinces Wall Street that the old rules of valuation no longer apply. Oklo was one of those stories. At our March summit, Joe Gamble looked past the story and looked at the business: Oklo builds and sells electricity. That makes it a utility, not a tech company, and utilities get valued on earnings. The average one trades around 20x profit. Oklo was sitting at a $9 billion valuation with no revenue and no earnings. NONE. And while investors bid the stock up, management was selling shares as fast as they could print them. That's the kind of thinking you get in this room. People who know the difference between a story and a balance sheet, and who say so before anyone else is willing to. 15 of them are doing it again on July 22 at our Best Stock Ideas Online Summit. One day, the single best ideas from elite investors that are actively playing the game. All for just $99. It's virtual with replays + transcripts available, grab your ticket today:

George Noble

16,841 次观看 • 1 个月前

Every Fortune 500 executive is buying AI subscriptions and calling it a strategy. Palantir CEO Alex Karp has a word for that. Karp: “The general approach of just buying models is going to be essentially self-pleasuring for an enterprise at the cost of the enterprise.” Karp: “You buy some large language model, you party with it basically, and the next day you have a hangover.” The entire corporate world is mispricing the AI transition. They are renting intelligence with no foundation to run it on. A raw model floating in a vacuum hallucinates over your unstructured data, generates the illusion of work, and executes nothing. The party ends. The hangover begins. Nothing changed. Karp identified exactly where the value actually goes. Karp: “All the value in the market is going to go to chips and what we call ontology.” Not the models. Not the subscriptions. Not the chatbot interfaces layered on top of them. The ontology. The precise digital architecture of how an organization actually operates. Its security permissions. Its supply chain physics. It’s operational logic. Karp: “The ontology will allow you to take a large language model and use it, refine it, and then impose it on your enterprise in the logic of your enterprise, in the security model of your enterprise.” When you bind a frontier model to the strict underlying logic of a specific enterprise, something fundamental shifts. It stops generating text. It starts generating action. Karp: “We’re using it on the battlefield, we’re using it to compress margins. We’re making engineers better engineers. We’re making people who are not engineers into engineers using our ontology and a large language model.” The traditional engineering bottleneck does not slow down. It disappears. Karp: “We are sitting on the only thing that actually creates quantifiable, transformational value.” The companies renting models are paying for the feeling of transformation. The companies building ontologies are executing the actual thing. One of them will define the next decade. The other will wake up in 2030 wondering where their market share went. Exactly like a hangover.

Every Fortune 500 executive is buying AI subscriptions and calling it a strategy. Palantir CEO Alex Karp has a word for that. Karp: “The general approach of just buying models is going to be essentially self-pleasuring for an enterprise at the cost of the enterprise.” Karp: “You buy some large language model, you party with it basically, and the next day you have a hangover.” The entire corporate world is mispricing the AI transition. They are renting intelligence with no foundation to run it on. A raw model floating in a vacuum hallucinates over your unstructured data, generates the illusion of work, and executes nothing. The party ends. The hangover begins. Nothing changed. Karp identified exactly where the value actually goes. Karp: “All the value in the market is going to go to chips and what we call ontology.” Not the models. Not the subscriptions. Not the chatbot interfaces layered on top of them. The ontology. The precise digital architecture of how an organization actually operates. Its security permissions. Its supply chain physics. It’s operational logic. Karp: “The ontology will allow you to take a large language model and use it, refine it, and then impose it on your enterprise in the logic of your enterprise, in the security model of your enterprise.” When you bind a frontier model to the strict underlying logic of a specific enterprise, something fundamental shifts. It stops generating text. It starts generating action. Karp: “We’re using it on the battlefield, we’re using it to compress margins. We’re making engineers better engineers. We’re making people who are not engineers into engineers using our ontology and a large language model.” The traditional engineering bottleneck does not slow down. It disappears. Karp: “We are sitting on the only thing that actually creates quantifiable, transformational value.” The companies renting models are paying for the feeling of transformation. The companies building ontologies are executing the actual thing. One of them will define the next decade. The other will wake up in 2030 wondering where their market share went. Exactly like a hangover.

Dustin

210,533 次观看 • 5 个月前

IN 1985 ONE OF THE GREATEST PHYSICISTS WHO EVER LIVED SAT DOWN TO EXPLAIN HOW COMPUTERS ACTUALLY WORK AND TOLD A ROOM FULL OF ENGINEERS THE MACHINE IS COMPLETELY DUMB 76 minutes from Richard Feynman, still called the clearest explanation of what a computer really is ever given. -> The idea that lands: a computer is just a very, very fast, very, very dumb file clerk. It doesn't think. It follows tiny simple rules, billions of times a second. All the complexity you're in awe of comes from stacking simple things. There's no magic underneath. There never was. Forty years later everyone calls the model "Intelligent". Feynman already told you what it really is: speed, not thought. Being amazed by the machine was never the point -> understanding what it's actually doing is. Most people are dazzled by what AI says. The ones who watched this know exactly what's happening underneath. Bookmark & Watch it today. This one's a legend ↓

IN 1985 ONE OF THE GREATEST PHYSICISTS WHO EVER LIVED SAT DOWN TO EXPLAIN HOW COMPUTERS ACTUALLY WORK AND TOLD A ROOM FULL OF ENGINEERS THE MACHINE IS COMPLETELY DUMB 76 minutes from Richard Feynman, still called the clearest explanation of what a computer really is ever given. -> The idea that lands: a computer is just a very, very fast, very, very dumb file clerk. It doesn't think. It follows tiny simple rules, billions of times a second. All the complexity you're in awe of comes from stacking simple things. There's no magic underneath. There never was. Forty years later everyone calls the model "Intelligent". Feynman already told you what it really is: speed, not thought. Being amazed by the machine was never the point -> understanding what it's actually doing is. Most people are dazzled by what AI says. The ones who watched this know exactly what's happening underneath. Bookmark & Watch it today. This one's a legend ↓

slash1s

259,903 次观看 • 1 个月前