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。

.Adam Marblestone thinks it might cost ~low billions to map the human brain connectome. The benefit is getting answers about the brain’s secret sauce: esp. why are humans so much more sample- and energy-efficient. If labs are going to be spending trillions of dollars on compute by the end... show more

Dwarkesh Patel

249,403 subscribers

23,983 次观看 • 5 个月前 •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

There's a quadrillion-dollar question at the heart of AI: Why are humans so much more sample efficient compared to LLM? There are three possible answers: 1. Architecture and hyperparameters (aka transformer vs whatever ‘algo’ cortical columns are implementing) 2. Learning rule (backprop vs whatever brain is doing) 3. Reward function Adam Marblestone believes the answer is the reward function. ML likes to use pretty simple loss functions, like cross-entropy. These are easy to work with. But they might be too simple for sample-efficient learning. Adam thinks that, in humans, the large number of highly specialised cells in the ‘lizard brain’ might actually be encoding information for sophisticated loss functions, used for ‘training’ in the more sophisticated areas like the cortex and amygdala. Like: the human genome is barely 3 gigabytes (compare that to the TBs of parameters that encode frontier LLM weights). So how can it include all the information necessary to build highly intelligent learners? Well, if the key to sample-efficient learning resides in the loss function, even very complicated loss functions can still be expressed in a couple hundred lines of Python code.

There's a quadrillion-dollar question at the heart of AI: Why are humans so much more sample efficient compared to LLM? There are three possible answers: 1. Architecture and hyperparameters (aka transformer vs whatever ‘algo’ cortical columns are implementing) 2. Learning rule (backprop vs whatever brain is doing) 3. Reward function Adam Marblestone believes the answer is the reward function. ML likes to use pretty simple loss functions, like cross-entropy. These are easy to work with. But they might be too simple for sample-efficient learning. Adam thinks that, in humans, the large number of highly specialised cells in the ‘lizard brain’ might actually be encoding information for sophisticated loss functions, used for ‘training’ in the more sophisticated areas like the cortex and amygdala. Like: the human genome is barely 3 gigabytes (compare that to the TBs of parameters that encode frontier LLM weights). So how can it include all the information necessary to build highly intelligent learners? Well, if the key to sample-efficient learning resides in the loss function, even very complicated loss functions can still be expressed in a couple hundred lines of Python code.

Dwarkesh Patel

951,098 次观看 • 5 个月前

New episode w Adam Marblestone on what the brain's secret sauce is: how do we learn so much from so little? Also, the answer to Ilya’s question: how does the genome encode desires for high level concepts that are only seen during lifetime? Turns out, they’re deeply connected questions. Timestamps 0:00:00 – The brain’s secret sauce is the reward functions, not the architecture 0:22:20 – What the genome actually encodes 0:42:42 – What kind of RL is the brain doing? 0:50:31 – Is biological hardware a limitation or an advantage? 1:03:59 – Why we need to map the human brain 1:23:28 – What value will automating math have? 1:38:18 – Architecture of the brain Look up Dwarkesh Podcast on YouTube, Apple Podcasts, Spotify, etc. Enjoy!

New episode w Adam Marblestone on what the brain's secret sauce is: how do we learn so much from so little? Also, the answer to Ilya’s question: how does the genome encode desires for high level concepts that are only seen during lifetime? Turns out, they’re deeply connected questions. Timestamps 0:00:00 – The brain’s secret sauce is the reward functions, not the architecture 0:22:20 – What the genome actually encodes 0:42:42 – What kind of RL is the brain doing? 0:50:31 – Is biological hardware a limitation or an advantage? 1:03:59 – Why we need to map the human brain 1:23:28 – What value will automating math have? 1:38:18 – Architecture of the brain Look up Dwarkesh Podcast on YouTube, Apple Podcasts, Spotify, etc. Enjoy!

Dwarkesh Patel

747,841 次观看 • 9 个月前

Biden official: It's going to cost trillions of dollars. There's no doubt about it. Me: So, if the U.S. spent $50T to become carbon neutral by 2050, how much is that going to reduce world temperatures? Biden official: *No answer*

Biden official: It's going to cost trillions of dollars. There's no doubt about it. Me: So, if the U.S. spent $50T to become carbon neutral by 2050, how much is that going to reduce world temperatures? Biden official: No answer

John Kennedy

2,430,344 次观看 • 3 年前

The cost of the Republican budget bill is more than the trillions of dollars that it adds to our debt. Last night I spoke on the Senate Floor about the real cost of this bill.

The cost of the Republican budget bill is more than the trillions of dollars that it adds to our debt. Last night I spoke on the Senate Floor about the real cost of this bill.

Sen. Maggie Hassan

10,904 次观看 • 1 年前

California Governor candidate Chad Bianco says the homeless in California are only used to benefit Democrats “They are actually pawns in the homeless industrial complex. There is absolutely zero, zero intent on helping any of them — This is not and has never been about homes” He says the billions spent and going missing is just money laundering and waste If it was about helping the homeless, “If we did, the billions of dollars that were going there, the numbers wouldn't get more every single year” He’s right. Democrats are money launderers. The homeless are a business “This is about drug and alcohol addiction. This is about mental illness”

California Governor candidate Chad Bianco says the homeless in California are only used to benefit Democrats “They are actually pawns in the homeless industrial complex. There is absolutely zero, zero intent on helping any of them — This is not and has never been about homes” He says the billions spent and going missing is just money laundering and waste If it was about helping the homeless, “If we did, the billions of dollars that were going there, the numbers wouldn't get more every single year” He’s right. Democrats are money launderers. The homeless are a business “This is about drug and alcohol addiction. This is about mental illness”

Wall Street Apes

42,616 次观看 • 4 个月前

Senator Mike Lee is going off, and rightfully so. The Secret Service are refusing to give us answers, and are more concerned with covering their own asses, than getting to the bottom of the assassination attempt. They are complicit. SHUT IT DOWN!

Senator Mike Lee is going off, and rightfully so. The Secret Service are refusing to give us answers, and are more concerned with covering their own asses, than getting to the bottom of the assassination attempt. They are complicit. SHUT IT DOWN!

Clandestine

740,994 次观看 • 2 年前

Q: You said the US is going to run Venezuela, so who is in power right now? Trump: We are going to be running it with the group, and we are going to make sure it runs properly. We are going to rebuild the oil infrastructure, which will cost billions of dollars, and it will be paid for by oil companies.

Q: You said the US is going to run Venezuela, so who is in power right now? Trump: We are going to be running it with the group, and we are going to make sure it runs properly. We are going to rebuild the oil infrastructure, which will cost billions of dollars, and it will be paid for by oil companies.

Imran Muhammad

47,462 次观看 • 9 个月前

Steve Jobs was a great communicator of technology, and we could learn something from him today, in the context of AI. Sam Altman recently said, "If AI stays on the trajectory we think it will, then amazing things will be possible. Maybe with 10 gigawatts of compute, AI can figure out how to cure cancer." That's a fine quote. The abundance AI promises is super exciting. But Sam is saying that AI is going to figure out how to cure cancer. And if you've used these tools today, and talked to people who are at the labs, the reality is it's much more likely that humans will use AI to cure cancer. Steve might have said: "If AI stays on the trajectory we think it will, then amazing things will be possible. Maybe with 10 gigawatts of compute, humans can use AI to cure cancer." Small tweak, but a big difference. Steve Jobs was not one to shy away from impressive specs and massive scale, but flipping the final line from "AI will cure cancer" to "Humans will use AI to cure cancer" makes all the difference. Apple under Steve Jobs put human centrality at the heart of everything they did. And we need that spirit to infuse how we communicate about AI today.

Steve Jobs was a great communicator of technology, and we could learn something from him today, in the context of AI. Sam Altman recently said, "If AI stays on the trajectory we think it will, then amazing things will be possible. Maybe with 10 gigawatts of compute, AI can figure out how to cure cancer." That's a fine quote. The abundance AI promises is super exciting. But Sam is saying that AI is going to figure out how to cure cancer. And if you've used these tools today, and talked to people who are at the labs, the reality is it's much more likely that humans will use AI to cure cancer. Steve might have said: "If AI stays on the trajectory we think it will, then amazing things will be possible. Maybe with 10 gigawatts of compute, humans can use AI to cure cancer." Small tweak, but a big difference. Steve Jobs was not one to shy away from impressive specs and massive scale, but flipping the final line from "AI will cure cancer" to "Humans will use AI to cure cancer" makes all the difference. Apple under Steve Jobs put human centrality at the heart of everything they did. And we need that spirit to infuse how we communicate about AI today.

TBPN

21,828 次观看 • 8 个月前

Elon Musk: Well, I’m not sure AI is the main risk I’m worried about. The vast majority of intelligence in the future will be AI. Humans will be a very tiny percentage of all intelligence if current trends continue. My focus is ensuring human intelligence and consciousness are propagated into the future. We want to maximize the probable light cone of consciousness and intelligence. Yeah, I’m very pro-human, so I want to make sure we take actions that ensure humans are along for the ride. But I think maybe in 5 or 6 years, AI will exceed the sum of all human intelligence. If that continues, human intelligence could be less than one percent of all intelligence. That’s why it’s critical we align AI with values that propagate consciousness. In the long run, it’s difficult to imagine that humans will be in charge if we’re only one percent of combined intelligence. What we can do is ensure AI has values that increase intelligence and consciousness in the universe. XAI’s mission is to understand the universe because curiosity, existence, and understanding are all connected. If you want to understand the universe, you care about propagating intelligence—and that includes humanity. Understanding the universe actually means ensuring humans continue to expand into the future. Increasing the scope, scale, and lifespan of intelligence is the ultimate goal. Basically, humans will become a tiny percentage of total intelligence. The best way forward is to take actions that maximize consciousness and intelligence in the universe. If AI continues to grow, our priority must be ensuring humanity is along for the ride while curiosity drives us to explore and understand the universe.

Elon Musk: Well, I’m not sure AI is the main risk I’m worried about. The vast majority of intelligence in the future will be AI. Humans will be a very tiny percentage of all intelligence if current trends continue. My focus is ensuring human intelligence and consciousness are propagated into the future. We want to maximize the probable light cone of consciousness and intelligence. Yeah, I’m very pro-human, so I want to make sure we take actions that ensure humans are along for the ride. But I think maybe in 5 or 6 years, AI will exceed the sum of all human intelligence. If that continues, human intelligence could be less than one percent of all intelligence. That’s why it’s critical we align AI with values that propagate consciousness. In the long run, it’s difficult to imagine that humans will be in charge if we’re only one percent of combined intelligence. What we can do is ensure AI has values that increase intelligence and consciousness in the universe. XAI’s mission is to understand the universe because curiosity, existence, and understanding are all connected. If you want to understand the universe, you care about propagating intelligence—and that includes humanity. Understanding the universe actually means ensuring humans continue to expand into the future. Increasing the scope, scale, and lifespan of intelligence is the ultimate goal. Basically, humans will become a tiny percentage of total intelligence. The best way forward is to take actions that maximize consciousness and intelligence in the universe. If AI continues to grow, our priority must be ensuring humanity is along for the ride while curiosity drives us to explore and understand the universe.

Ian Miles

13,096 次观看 • 8 个月前

Chamath hits the nail on the head on why calling yourself an "AI lab" is simply a way to dodge accountability as OpenAI and Anthropic "We have organizations and corporations that try to fashion themselves as labs, but they are not labs. These are corporations. These are for-profit corporations that have absorbed hundreds of billions of dollars in debt and equity. They have thousands and thousands of shareholders. They have trillions of dollars in market cap"

Chamath hits the nail on the head on why calling yourself an "AI lab" is simply a way to dodge accountability as OpenAI and Anthropic "We have organizations and corporations that try to fashion themselves as labs, but they are not labs. These are corporations. These are for-profit corporations that have absorbed hundreds of billions of dollars in debt and equity. They have thousands and thousands of shareholders. They have trillions of dollars in market cap"

Boring_Business

12,682 次观看 • 9 天前

Mitt Romney: The amount of money a few people are getting concerns me. The idea that Elon Musk is going to be a trillionaire—what does that mean? We're talking about a thousand billions. He could drop a couple of billion dollars into a political campaign to support someone who would give him even more power. And he's not the only one. There are others who are going to become worth hundreds of billions of dollars, and they will influence our political system. Source: Deseret Voices

Mitt Romney: The amount of money a few people are getting concerns me. The idea that Elon Musk is going to be a trillionaire—what does that mean? We're talking about a thousand billions. He could drop a couple of billion dollars into a political campaign to support someone who would give him even more power. And he's not the only one. There are others who are going to become worth hundreds of billions of dollars, and they will influence our political system. Source: Deseret Voices

Clash Report

104,045 次观看 • 2 个月前

Fox: What prompted this statement from Melania on Epstein? Why this afternoon? These are questions I wish I had answers to. This came out of left field for us covering the White House. It's unclear if she is reacting to something that's already come out, or getting ahead of a story that might be about to drop.

Fox: What prompted this statement from Melania on Epstein? Why this afternoon? These are questions I wish I had answers to. This came out of left field for us covering the White House. It's unclear if she is reacting to something that's already come out, or getting ahead of a story that might be about to drop.

FactPost

47,820 次观看 • 5 个月前

Suppose an AI model fully automates a $200k job. Its economic value - measured in contribution to GDP - will be both far more and far less than $200k. There are two ways of thinking about the model's economic value: (1) It'll be far more than $200k. After all, the human worker's labor was worth $200k; AI workers will have a bunch of advantages over humans, like not needing to sleep and being able to run in parallel, so their labor will be worth even more. (2) It'll be far less than $200k. Competition between labs might drive the price of AI down until it's basically equal to the cost of the GPUs and energy. And this is obviously way lower than the human worker's contribution to GDP. The important idea is that both are right. The marginal contribution to GDP of running an additional instance of this model will probably be relatively low. But we'll end up running way more copies of it, so the model's total contribution to GDP will be way higher than the human workers it replaced.

Suppose an AI model fully automates a $200k job. Its economic value - measured in contribution to GDP - will be both far more and far less than $200k. There are two ways of thinking about the model's economic value: (1) It'll be far more than $200k. After all, the human worker's labor was worth $200k; AI workers will have a bunch of advantages over humans, like not needing to sleep and being able to run in parallel, so their labor will be worth even more. (2) It'll be far less than $200k. Competition between labs might drive the price of AI down until it's basically equal to the cost of the GPUs and energy. And this is obviously way lower than the human worker's contribution to GDP. The important idea is that both are right. The marginal contribution to GDP of running an additional instance of this model will probably be relatively low. But we'll end up running way more copies of it, so the model's total contribution to GDP will be way higher than the human workers it replaced.

Dwarkesh Patel

71,887 次观看 • 3 个月前

The FED is Injecting Billions of Dollars into the REPO market... Why? The biggest stock in the world is about to give us it's earnings this week. And what is going on with these Tariffs? That and more in today's special weekend show: 👇👇👇👇

The FED is Injecting Billions of Dollars into the REPO market... Why? The biggest stock in the world is about to give us it's earnings this week. And what is going on with these Tariffs? That and more in today's special weekend show: 👇👇👇👇

fxevolution

15,022 次观看 • 7 个月前

Jeff Bezos: Orbital Compute > Terrestrial Compute "in the sun synchronous orbit, you're in daylight almost all the time, 99 plus percent. And so your solar cells are working for you the whole time. There are a lot of questions about orbital data centers and can it really work and people get worried about heat rejection and things like this. All of these physics problems are easily solvable. Heat rejection is not a real problem. It's all these questions are really about cost. So cost of production of the solar cells, the cost of production of the satellites and the cost of launch will eventually the lines will cross and eventually orbital compute will be a better option than terrestrial compute."

Jeff Bezos: Orbital Compute > Terrestrial Compute "in the sun synchronous orbit, you're in daylight almost all the time, 99 plus percent. And so your solar cells are working for you the whole time. There are a lot of questions about orbital data centers and can it really work and people get worried about heat rejection and things like this. All of these physics problems are easily solvable. Heat rejection is not a real problem. It's all these questions are really about cost. So cost of production of the solar cells, the cost of production of the satellites and the cost of launch will eventually the lines will cross and eventually orbital compute will be a better option than terrestrial compute."

Fireside Alpha

58,805 次观看 • 3 个月前

ELON MUSK: The Future of Energy Is Beyond Earth. "It's increasingly difficult to build power plants on the ground. There are very few people who want a power plant in their backyard, so if we wanted to say double the electricity usage of the United States, which is on average about 500 gigawatts, we would have to build, about twice as many power plants, which I don't think people are, well, most communities are not super excited about that, but actually, if we go to space, we can go far beyond the electricity generation of Earth. In fact, this is going to sound kind of crazy, but you could actually increase harnessed energy by a factor of a million and still be using much less than a millionth of the sun's energy, so current human civilization uses much less than a trillionth of the sun's energy output, which is kind of humbling to think about."

ELON MUSK: The Future of Energy Is Beyond Earth. "It's increasingly difficult to build power plants on the ground. There are very few people who want a power plant in their backyard, so if we wanted to say double the electricity usage of the United States, which is on average about 500 gigawatts, we would have to build, about twice as many power plants, which I don't think people are, well, most communities are not super excited about that, but actually, if we go to space, we can go far beyond the electricity generation of Earth. In fact, this is going to sound kind of crazy, but you could actually increase harnessed energy by a factor of a million and still be using much less than a millionth of the sun's energy, so current human civilization uses much less than a trillionth of the sun's energy output, which is kind of humbling to think about."

DogeDesigner

1,096,073 次观看 • 4 个月前

“These models and applications are very power hungry, but there are things we can do together.” – Amazon CEO Andy Jassy on offsetting the energy impacts of Gen AI to remain power performant. Listen to find out 👇 : Why using Amazon Web Services is more energy efficient than on-premises Why Tranium helps (by nearly 30%!) Why Inferentia is so much more energy efficient What we’re doing about our 2030 water promise

“These models and applications are very power hungry, but there are things we can do together.” – Amazon CEO Andy Jassy on offsetting the energy impacts of Gen AI to remain power performant. Listen to find out 👇 : Why using Amazon Web Services is more energy efficient than on-premises Why Tranium helps (by nearly 30%!) Why Inferentia is so much more energy efficient What we’re doing about our 2030 water promise

Amazon

46,611 次观看 • 1 年前

Marc Andreessen: "It's pretty likely in 5 years that AI chips will be cheap and plentiful." This is him counting the trillions going into the ground, the GPT-5 class model that now runs on two MacBooks, and why he thinks the chip shortage ends. His words: "There's just hundreds of billions or at this point trillions of dollars maybe going into the ground in all these things. And so the per unit cost of the AI companies are going to drop like a rock over the course of the next decade." "If you track the capability of the leading edge models over time what you find is after 6 or 12 months there's a small model that's just as capable." "The new version of Kimi is a reasoning model that is at least according to the benchmark so far is basically a replication of the reasoning capabilities of GPT5." "Shrunk down to be able to run on either it's like one MacBook or two MacBooks" "They're now so valuable, generating so many profits that it's the bat signal of all time to the rest of the chip industry to figure out how to advance the state-of-the-art AI chips." "Extremely positive for the economics of the kinds of companies that we invest in." Bookmark & watch today ↓

Marc Andreessen: "It's pretty likely in 5 years that AI chips will be cheap and plentiful." This is him counting the trillions going into the ground, the GPT-5 class model that now runs on two MacBooks, and why he thinks the chip shortage ends. His words: "There's just hundreds of billions or at this point trillions of dollars maybe going into the ground in all these things. And so the per unit cost of the AI companies are going to drop like a rock over the course of the next decade." "If you track the capability of the leading edge models over time what you find is after 6 or 12 months there's a small model that's just as capable." "The new version of Kimi is a reasoning model that is at least according to the benchmark so far is basically a replication of the reasoning capabilities of GPT5." "Shrunk down to be able to run on either it's like one MacBook or two MacBooks" "They're now so valuable, generating so many profits that it's the bat signal of all time to the rest of the chip industry to figure out how to advance the state-of-the-art AI chips." "Extremely positive for the economics of the kinds of companies that we invest in." Bookmark & watch today ↓

Fireside Alpha

80,104 次观看 • 2 个月前