正在加载视频...

视频加载失败

返回首页

Another piece from the Sasha Rush conversation, this time on rewards for coding RL. He said Cursor uses a mix. Some rewards look at the tool calls themselves, some only at the final output. Everything end-to-end, no process rewards guessing what happens in the middle. I agree with him...

ziv_ravid's profile picture

Ravid Shwartz Ziv

12,050 subscribers

14,826 次观看 • 2 个月前 •via X (Twitter)

LIVE

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

Q: So this second place is enough for you and Ferrari? "I think in true performance this is where this is great you know. Ultimately, the two Mercedes should be ahead of us. So, to get to the second, and in the end would have been a third as well, there would have been really good points for the team, but the conditions were tricky out there so...." "That's why I want to thank everyone back at the factory. We have to keep pushing, because its only the beginning of the season and we're right there, we're right there... so we're going to hunt these guys down."
0:36

Q: So this second place is enough for you and Ferrari? "I think in true performance this is where this is great you know. Ultimately, the two Mercedes should be ahead of us. So, to get to the second, and in the end would have been a third as well, there would have been really good points for the team, but the conditions were tricky out there so...." "That's why I want to thank everyone back at the factory. We have to keep pushing, because its only the beginning of the season and we're right there, we're right there... so we're going to hunt these guys down."

sim 🇧🇷🇲🇽🇸🇳🇫🇷🇭🇹

22,297 次观看 • 1 个月前

I have a friend who doesn't read anything published in the past 50 years, and the more I think about it, the more I think he's onto something. The reason is that time is the best filter we have for quality. People are bad at judging quality in the moment but very good at getting rid of junk over time. — — "History is not very good at capturing all that is great in art. It is not good at that. There are many great symphonies that have been lost permanently, there are many great painters that died unknown and their paintings are gone, there's novels that have been written that no one will ever read. So history is not good at capturing all this great art. But history is very good at discarding all that is mediocre. And the amount of time that that takes, it's something like 50 years. So over the course of 50 years, what will happen is a lot of stuff that was prominent will be re-filtered and re-filtered and re-filtered, and you'll end up with a smaller group of things which have survived that test of time. So if you think about it right now, if you go back and look at the bestseller lists for 1974, 1973, there's a lot of that that would have been highly regarded at the time, which people do not read anymore for a variety of reasons, and there's some that has survived, and that's a very telling distinction. So in a world where I'm turning 60 this year, you have a limited amount of time, all four of us have active lives, we want to make sure that if we're going to sit down, we're going to read carefully, we're going to meet and we're going to discuss it in detail, we want to make sure that the work is rewarding. And the best way to ensure that is by drawing from the past." amor towles
1:31

I have a friend who doesn't read anything published in the past 50 years, and the more I think about it, the more I think he's onto something. The reason is that time is the best filter we have for quality. People are bad at judging quality in the moment but very good at getting rid of junk over time. — — "History is not very good at capturing all that is great in art. It is not good at that. There are many great symphonies that have been lost permanently, there are many great painters that died unknown and their paintings are gone, there's novels that have been written that no one will ever read. So history is not good at capturing all this great art. But history is very good at discarding all that is mediocre. And the amount of time that that takes, it's something like 50 years. So over the course of 50 years, what will happen is a lot of stuff that was prominent will be re-filtered and re-filtered and re-filtered, and you'll end up with a smaller group of things which have survived that test of time. So if you think about it right now, if you go back and look at the bestseller lists for 1974, 1973, there's a lot of that that would have been highly regarded at the time, which people do not read anymore for a variety of reasons, and there's some that has survived, and that's a very telling distinction. So in a world where I'm turning 60 this year, you have a limited amount of time, all four of us have active lives, we want to make sure that if we're going to sit down, we're going to read carefully, we're going to meet and we're going to discuss it in detail, we want to make sure that the work is rewarding. And the best way to ensure that is by drawing from the past." amor towles

David Perell

86,741 次观看 • 1 年前

You know the thing with random rewards? The thing is that you never get what you want, right? And everyone is hoping for Manga Kenji Skin now... but is it really everyone? What does your heart most desire? If you look back at your life journey, are you where you wanted to be? And if not, why not? What changes would you have made in your life to be happy with yourself? Because this is what really matters. Your reaction to this reward is only a reflection of how deeply you feel inside, isn't it? Is the reward good? Is the reward bad? Isn't that only a matter of perspective? ANYWAY, enough talking, let's finally announce the reward you will receive today, after you watch this video, and after you read this text... But, what is the point of having the reward written here in this text, if it's also announced in the video AND announced in the game... isn't all of this pointless? What does it matter what door you open in the end, right? Some of them have hints, some of them don't, it's pretty much random. OR IS IT? The Mega Box door was pretty obvious, wasn't it? I guess some of them DO have hints, and some of them don't. We did say that in the video actually... the trick is to find out which ones are misleading and which ones are not. To be honest, the plain white door wasn't designed with Manga Kenji in mind... you were the ones who came up with your own theory and when it didn't land, you thought we fooled you, but in reality, you fooled yourself. But one thing I can say, we've seen a lot of crazy theories out there, and some of them are actually right, surprisingly. Maybe an accident, maybe whoever came up with that theory is a future version of me. It can happen, I read it in a book once. Or maybe not. ANYWAY, hope you have enjoyed today's reward, and let's bring the community together to vote for the best door today! #ScaryDoors
2:14

You know the thing with random rewards? The thing is that you never get what you want, right? And everyone is hoping for Manga Kenji Skin now... but is it really everyone? What does your heart most desire? If you look back at your life journey, are you where you wanted to be? And if not, why not? What changes would you have made in your life to be happy with yourself? Because this is what really matters. Your reaction to this reward is only a reflection of how deeply you feel inside, isn't it? Is the reward good? Is the reward bad? Isn't that only a matter of perspective? ANYWAY, enough talking, let's finally announce the reward you will receive today, after you watch this video, and after you read this text... But, what is the point of having the reward written here in this text, if it's also announced in the video AND announced in the game... isn't all of this pointless? What does it matter what door you open in the end, right? Some of them have hints, some of them don't, it's pretty much random. OR IS IT? The Mega Box door was pretty obvious, wasn't it? I guess some of them DO have hints, and some of them don't. We did say that in the video actually... the trick is to find out which ones are misleading and which ones are not. To be honest, the plain white door wasn't designed with Manga Kenji in mind... you were the ones who came up with your own theory and when it didn't land, you thought we fooled you, but in reality, you fooled yourself. But one thing I can say, we've seen a lot of crazy theories out there, and some of them are actually right, surprisingly. Maybe an accident, maybe whoever came up with that theory is a future version of me. It can happen, I read it in a book once. Or maybe not. ANYWAY, hope you have enjoyed today's reward, and let's bring the community together to vote for the best door today! #ScaryDoors

Brawl Stars

315,498 次观看 • 8 个月前

Subramanyam: Well, the reality is, you know, this administration works for Donald Trump, right? Even the independent agencies and even prosecutors who are supposed to put law and justice above politics have been told to only investigate Democrats and to only do what makes the president look good throughout this entire investigation. So if they are in possession of the files, we have deep concerns that they would manipulate those files in some way. The other thing we are concerned they’re going to do is only release files that either make the president look good or only release files that the president chooses to agree to release — and not tell the whole story, protecting the president and his friends, and anyone else the president deems worth protecting. So, this is a concern. He’s signed the bill, but this is only the beginning of the fight to make sure we get the full, unredacted release of the files, which is what we really need.
1:01

Subramanyam: Well, the reality is, you know, this administration works for Donald Trump, right? Even the independent agencies and even prosecutors who are supposed to put law and justice above politics have been told to only investigate Democrats and to only do what makes the president look good throughout this entire investigation. So if they are in possession of the files, we have deep concerns that they would manipulate those files in some way. The other thing we are concerned they’re going to do is only release files that either make the president look good or only release files that the president chooses to agree to release — and not tell the whole story, protecting the president and his friends, and anyone else the president deems worth protecting. So, this is a concern. He’s signed the bill, but this is only the beginning of the fight to make sure we get the full, unredacted release of the files, which is what we really need.

Acyn

46,383 次观看 • 7 个月前

Gabbit Robotics co-founder Nicole Mattero on why the next AI winners won't fight over your screen, they'll fight to get closer to you: She kicks things off with what she admits is a hot take: "My hot take is that hardware is the only moat going forward because anyone's going to be able to spin up a model and make it really good and really cheap." Her logic is simple. Once a strong model is cheap and easy for anyone to build, the model alone can't set you apart. So what actually protects you? "Where does that leave you? Basically companies that have tactile modes such as hardware are the only ones that actually survive." Hardware, though, is only one piece of her argument. The second piece is data: "The models that win in the long run are the ones where private data is their [moat]." From there, Mattero ties the two together. Hardware gives a company one moat. The ability to gather unique personal data gives it another. Put both in place and the edge starts to multiply: "When you have a company that has hardware as a [moat], but then B is able to collect really unique personal data, that also becomes a [moat]." And that, she says, is the reason so many foundation model companies are reaching beyond the screen and into the physical world: "That's why I think a lot of these foundation models are like, let's get something in the home or on the person. It gives them an edge on both fronts."
0:45

Gabbit Robotics co-founder Nicole Mattero on why the next AI winners won't fight over your screen, they'll fight to get closer to you: She kicks things off with what she admits is a hot take: "My hot take is that hardware is the only moat going forward because anyone's going to be able to spin up a model and make it really good and really cheap." Her logic is simple. Once a strong model is cheap and easy for anyone to build, the model alone can't set you apart. So what actually protects you? "Where does that leave you? Basically companies that have tactile modes such as hardware are the only ones that actually survive." Hardware, though, is only one piece of her argument. The second piece is data: "The models that win in the long run are the ones where private data is their [moat]." From there, Mattero ties the two together. Hardware gives a company one moat. The ability to gather unique personal data gives it another. Put both in place and the edge starts to multiply: "When you have a company that has hardware as a [moat], but then B is able to collect really unique personal data, that also becomes a [moat]." And that, she says, is the reason so many foundation model companies are reaching beyond the screen and into the physical world: "That's why I think a lot of these foundation models are like, let's get something in the home or on the person. It gives them an edge on both fronts."

Big Brain AI

19,195 次观看 • 1 个月前

Satya Nadella says the next AI interface is not chat. It is the control room for 100 agents working overnight under your delegated authority. "Coding has worked so well that we now have to rebuild the IDE." "Oh my god, I have these 100 agent sessions. The cognitive load it transfers back to me as a human is so excessive that now I need a new UI." "The chat as the only artifact is also impossible. So that's why we need a canvas." Then he gives the enterprise version: "All through the night there was a bunch of stuff that all these autopilots that I have working on my behalf with my delegated authority did." And the operator question becomes: "What did you do? Did I do this work?" The real bottleneck is not whether agents can generate more tasks. It is whether humans get the interface, memory, permissions, and audit trail to safely manage 100 delegated workers.
2:09

Satya Nadella says the next AI interface is not chat. It is the control room for 100 agents working overnight under your delegated authority. "Coding has worked so well that we now have to rebuild the IDE." "Oh my god, I have these 100 agent sessions. The cognitive load it transfers back to me as a human is so excessive that now I need a new UI." "The chat as the only artifact is also impossible. So that's why we need a canvas." Then he gives the enterprise version: "All through the night there was a bunch of stuff that all these autopilots that I have working on my behalf with my delegated authority did." And the operator question becomes: "What did you do? Did I do this work?" The real bottleneck is not whether agents can generate more tasks. It is whether humans get the interface, memory, permissions, and audit trail to safely manage 100 delegated workers.

Karl Mehta

22,164 次观看 • 19 天前

Gasparino: You do have to ask yourself why the rush to get a deal like this done.. They didn’t agree to that much for us. Why did they go from DEFCON 1 to this? One reason is that there is a federal appeals court that is ruling on Trump’s use of the emergency power to impose tariffs and that court could rule against him. If it rules against him, the whole tariff scheme is up in the air
0:52

Gasparino: You do have to ask yourself why the rush to get a deal like this done.. They didn’t agree to that much for us. Why did they go from DEFCON 1 to this? One reason is that there is a federal appeals court that is ruling on Trump’s use of the emergency power to impose tariffs and that court could rule against him. If it rules against him, the whole tariff scheme is up in the air

Acyn

64,196 次观看 • 11 个月前

"The appropriate response from the Somali community would be to come out and apologize to the public and take accountability, and at least having some shame for doing that." You would think so, right? Not in this case. These people have no shame and will deny and play the victim until the bitter end. This is not only a cultural problem with these Somalians, it is a religious issue. Islam teaches that it is ok to do things like this so long as it is in the name of Islam. What did we expect when we imported low IQ people from the third world?
0:28

"The appropriate response from the Somali community would be to come out and apologize to the public and take accountability, and at least having some shame for doing that." You would think so, right? Not in this case. These people have no shame and will deny and play the victim until the bitter end. This is not only a cultural problem with these Somalians, it is a religious issue. Islam teaches that it is ok to do things like this so long as it is in the name of Islam. What did we expect when we imported low IQ people from the third world?

AmericanPapaBear™

23,170 次观看 • 6 个月前

Chamath: US AI Startups Can Learn A Lot from DeepSeek "This is a case where necessity was the mother of invention." On E213, Chamath Palihapitiya explained how, even if the $6M number is inaccurate, DeepSeek still had some impressive breakthroughs: GRPO > PPO for reinforcement learning "These guys were like, 'Well, how am I going to do this reinforcement learning thing?' They invented a totally different algorithm." "It uses a lot less computer memory and it's highly performant." PTX > CUDA to build the model "And then the second thing that was crazy is everybody is used to building models and compiling through CUDA, which is NVIDIA's proprietary language." "And these guys worked totally around CUDA and they did something called PTX, which goes right to the bare metal." What US AI startups can learn from this: "We, meaning the West, with all the money that we've had, didn't come up with these ideas." "And I think part of why we didn't come up is not that we're not smart enough to do it, but we weren't forced to because the constraints didn't exist." "And so I just wonder how we make sure we learn this principle, meaning, when the AI company wakes up and rolls out of bed and some VC gives them $200M, maybe that's not the right answer for a Series A or a Seed." "And maybe the right answer is $2M so that they do these DeepSeek-like innovations."
0:55

Chamath: US AI Startups Can Learn A Lot from DeepSeek "This is a case where necessity was the mother of invention." On E213, Chamath Palihapitiya explained how, even if the $6M number is inaccurate, DeepSeek still had some impressive breakthroughs: GRPO > PPO for reinforcement learning "These guys were like, 'Well, how am I going to do this reinforcement learning thing?' They invented a totally different algorithm." "It uses a lot less computer memory and it's highly performant." PTX > CUDA to build the model "And then the second thing that was crazy is everybody is used to building models and compiling through CUDA, which is NVIDIA's proprietary language." "And these guys worked totally around CUDA and they did something called PTX, which goes right to the bare metal." What US AI startups can learn from this: "We, meaning the West, with all the money that we've had, didn't come up with these ideas." "And I think part of why we didn't come up is not that we're not smart enough to do it, but we weren't forced to because the constraints didn't exist." "And so I just wonder how we make sure we learn this principle, meaning, when the AI company wakes up and rolls out of bed and some VC gives them $200M, maybe that's not the right answer for a Series A or a Seed." "And maybe the right answer is $2M so that they do these DeepSeek-like innovations."

The All-In Podcast

108,815 次观看 • 1 年前

Recently met Sasha Rush and he started giving me an impromptu lecture on how targeted on-policy self-distillation works. I asked him if I could record it on my iPhone. The basic idea is this: if the model made a mistake at some point in the rollout (for example, calling a tool that doesn't exist), we want to discourage this specific error, but we don't want to just learn from the final reward, because it's a very noisy signal spread out over the whole trajectory. So we have another model read this trajectory and figure where the error was made. It simply inserts some hint tokens to the part of the trajectory right above where the mistake was made. Now with these injected hint tokens, have the model run a forward pass. You're not having to regenerate a new rollout - aka no new decode required. The hint causes the model to assign lower probabilities to the error tokens. You then trains the original model to match these new probabilities, teaching it to downweight that specific mistake.
13:14

Recently met Sasha Rush and he started giving me an impromptu lecture on how targeted on-policy self-distillation works. I asked him if I could record it on my iPhone. The basic idea is this: if the model made a mistake at some point in the rollout (for example, calling a tool that doesn't exist), we want to discourage this specific error, but we don't want to just learn from the final reward, because it's a very noisy signal spread out over the whole trajectory. So we have another model read this trajectory and figure where the error was made. It simply inserts some hint tokens to the part of the trajectory right above where the mistake was made. Now with these injected hint tokens, have the model run a forward pass. You're not having to regenerate a new rollout - aka no new decode required. The hint causes the model to assign lower probabilities to the error tokens. You then trains the original model to match these new probabilities, teaching it to downweight that specific mistake.

Dwarkesh Patel

425,580 次观看 • 1 个月前

Catherine Austin Fitts argues that Donald Trump was installed by bankers to construct a digital control grid comprised of programmable money, digital ID, and a social credit system. She maintains that Trump is expendable now that the infrastructure for this system is complete. "So let's look at the control grid. There are three general baskets of what you need to do. The first is programmable money. The second is digital ID... And then the third is you need the hardware and the software infrastructure to do the social credit system." "And so you need all the data centres... But you also need a surveillance infrastructure and enforcement infrastructure to back it." "So how do you persuade conservatives... to embrace building that infrastructure? The way you do it is you say, oh, we have election fraud, or oh, we have an immigration problem. And so we need a digital ID to identify who everybody is so we can't have fraud." "If you look at everything [Trump has] done to build the control grid, or to put the conditions in place to do it, they now have what they need." "Trump is basically expendable now."
2:40

Catherine Austin Fitts argues that Donald Trump was installed by bankers to construct a digital control grid comprised of programmable money, digital ID, and a social credit system. She maintains that Trump is expendable now that the infrastructure for this system is complete. "So let's look at the control grid. There are three general baskets of what you need to do. The first is programmable money. The second is digital ID... And then the third is you need the hardware and the software infrastructure to do the social credit system." "And so you need all the data centres... But you also need a surveillance infrastructure and enforcement infrastructure to back it." "So how do you persuade conservatives... to embrace building that infrastructure? The way you do it is you say, oh, we have election fraud, or oh, we have an immigration problem. And so we need a digital ID to identify who everybody is so we can't have fraud." "If you look at everything [Trump has] done to build the control grid, or to put the conditions in place to do it, they now have what they need." "Trump is basically expendable now."

Wide Awake Media

148,643 次观看 • 3 个月前

DeepMind CEO Demis Hassabis says the AI bottleneck is already moving from model ideas to physical supply chains. "How much of the restriction at this point is physical? I am talking memory shortage, GPU shortage, and electricity." Demis: "That is constraining a lot of the deployment for sure. We see so much more demand for our models - Gemini and the other models that we have - than we can possibly serve." Then the less obvious part: "It does constrain a little bit the research, because you need a lot of chips to be able to experiment on new ideas at big enough scale that you can actually see if they are going to work." And on memory: "The whole supply chain is kind of strained... It still in the end actually comes down to a few suppliers of a few of the key components. Anywhere where there is some kind of constraint on the capacity, there is a sort of choke point." Important framing: The next AI race is not just who has the best model architecture. It is who can secure memory, chips, power, suppliers, and enough experimental capacity to find the next breakthrough before everyone else.
1:09

DeepMind CEO Demis Hassabis says the AI bottleneck is already moving from model ideas to physical supply chains. "How much of the restriction at this point is physical? I am talking memory shortage, GPU shortage, and electricity." Demis: "That is constraining a lot of the deployment for sure. We see so much more demand for our models - Gemini and the other models that we have - than we can possibly serve." Then the less obvious part: "It does constrain a little bit the research, because you need a lot of chips to be able to experiment on new ideas at big enough scale that you can actually see if they are going to work." And on memory: "The whole supply chain is kind of strained... It still in the end actually comes down to a few suppliers of a few of the key components. Anywhere where there is some kind of constraint on the capacity, there is a sort of choke point." Important framing: The next AI race is not just who has the best model architecture. It is who can secure memory, chips, power, suppliers, and enough experimental capacity to find the next breakthrough before everyone else.

Karl Mehta

25,716 次观看 • 18 天前

THE MOST IMPORTANT Q&A OF MEDIA DAY. Mariana: You came from a very solid weekend on top of everything, but at the same time, it seems that you don't feel that the team is listening to you. Am I right? And how do you balance that? Lewis: I feel like we're going in the right direction. Rome wasn't built in one day, so it takes time to build. For me, coming into the team, I wanted to be respectful of the way they've done things in the past and just to really observe and see where our strengths and where our weaknesses are and to highlight where our weaknesses are and areas that we need to work on. But I do feel that they've been responding. I think you're starting to see, hopefully, some of the impact of the work that we're doing in the background and also into next year's car. This is a car that I've had nothing to do with in terms of developing this car over the years. Hopefully, from next year, my input goes into that car, and that will be a car that I've hopefully been a part of or will have been a part of developing. But I think we've got a really great rapport. I think we're really progressing, particularly since the summer break. I think things have started to get better, and it's all just about building trust and communication. Also, I'm coming into a team that English is not the first language, and I don't speak Italian, so it's finding a common ground. And the fact is we all want to win. We're all here to achieve the same thing, and we've got to just keep pushing. So that's why I'm trying to keep everyone motivated on difficult weekends, trying to keep everyone lifted up. But there have been many, many things we've changed this year that I suggested that they hadn't done in the past, and so they have been listening. It doesn't change straight away, just like that. It takes time to build. And as engineers, they really need proof. They need numbers. That's what they work on. So you have to sometimes push to get certain changes to be made, and then when you change it and then it works, you're like, okay. Mariana: That's what I was talking about.. Lewis: Yeah! - F1 2025 Mexico -
2:28

THE MOST IMPORTANT Q&A OF MEDIA DAY. Mariana: You came from a very solid weekend on top of everything, but at the same time, it seems that you don't feel that the team is listening to you. Am I right? And how do you balance that? Lewis: I feel like we're going in the right direction. Rome wasn't built in one day, so it takes time to build. For me, coming into the team, I wanted to be respectful of the way they've done things in the past and just to really observe and see where our strengths and where our weaknesses are and to highlight where our weaknesses are and areas that we need to work on. But I do feel that they've been responding. I think you're starting to see, hopefully, some of the impact of the work that we're doing in the background and also into next year's car. This is a car that I've had nothing to do with in terms of developing this car over the years. Hopefully, from next year, my input goes into that car, and that will be a car that I've hopefully been a part of or will have been a part of developing. But I think we've got a really great rapport. I think we're really progressing, particularly since the summer break. I think things have started to get better, and it's all just about building trust and communication. Also, I'm coming into a team that English is not the first language, and I don't speak Italian, so it's finding a common ground. And the fact is we all want to win. We're all here to achieve the same thing, and we've got to just keep pushing. So that's why I'm trying to keep everyone motivated on difficult weekends, trying to keep everyone lifted up. But there have been many, many things we've changed this year that I suggested that they hadn't done in the past, and so they have been listening. It doesn't change straight away, just like that. It takes time to build. And as engineers, they really need proof. They need numbers. That's what they work on. So you have to sometimes push to get certain changes to be made, and then when you change it and then it works, you're like, okay. Mariana: That's what I was talking about.. Lewis: Yeah! - F1 2025 Mexico -

sim

170,303 次观看 • 8 个月前

#WATCH | US President Donald Trump says, "The war (Russia-Ukraine) is going to end. When it ends, I can't tell you but the war is going to end and this gentleman wants it to end and Vladimir Putin wants it to end. I think the whole world is tired of it. We are going to get it ended. I have ended 6 wars and I thought that maybe this would be the easiest one. And it's not the easiest one. It's a tough one...India-Pakistan, we are talking about big places. You take a look at some of these wars, you go to Africa and take a look at that. Rawanda and the Congo - that has been going on for 31 years. We have done a total of 6, not including the fact that we totally obliterated the future nuclear capability of Iran...I feel confident that we are going to get this war off." (Video: White House)
1:25

#WATCH | US President Donald Trump says, "The war (Russia-Ukraine) is going to end. When it ends, I can't tell you but the war is going to end and this gentleman wants it to end and Vladimir Putin wants it to end. I think the whole world is tired of it. We are going to get it ended. I have ended 6 wars and I thought that maybe this would be the easiest one. And it's not the easiest one. It's a tough one...India-Pakistan, we are talking about big places. You take a look at some of these wars, you go to Africa and take a look at that. Rawanda and the Congo - that has been going on for 31 years. We have done a total of 6, not including the fact that we totally obliterated the future nuclear capability of Iran...I feel confident that we are going to get this war off." (Video: White House)

ANI

105,010 次观看 • 11 个月前

BREAKING.🚨 Judge Merchan has instructed the jury they do not need to have a *UNANIMOUS* verdict in order to convict former President Donald J. Trump. "One thing in particular that the judge said the jurors could do. He delivered what is being called really the pinnacle of all of this. There is no need to agree on what has occurred. They can disagree on what the crime was among the three choices." "This means they could split 4-4-4 and the judge would still treat them unanimously. What does that mean?" "Outrageous. In a normal criminal case every statutory crime has what we call elements of the offense. Like in a bank robbery case you have to rob – it has to be a financial institution, you have to show intent," said former prosecutor Andrew McCarthy. "Those are the things the jury has to agree on unanimously that they were proved beyond a reasonable doubt. Here what we’re doing is taking the element that actually makes this a felony, because remember falsification of records is normally a misdemeanor in New York. What makes it a felony is that you are concealing or committing another crime." "And here the judge is telling them they don’t have to agree about what the other crime is under circumstances where that not only is what makes this a felony, makes it a four-year potential prison penalty rather than a year or less, but it is also what gets us into the courtroom." "If this had been a misdemeanor, the time to bring this case would have lapsed in 2019. The only reason they are still able to bring this case is because it’s a felony allegedly and yet now the judge is saying you know, you don’t have to agree on what the felony is." The jury has now gone to deliberations.
1:45

BREAKING.🚨 Judge Merchan has instructed the jury they do not need to have a *UNANIMOUS* verdict in order to convict former President Donald J. Trump. "One thing in particular that the judge said the jurors could do. He delivered what is being called really the pinnacle of all of this. There is no need to agree on what has occurred. They can disagree on what the crime was among the three choices." "This means they could split 4-4-4 and the judge would still treat them unanimously. What does that mean?" "Outrageous. In a normal criminal case every statutory crime has what we call elements of the offense. Like in a bank robbery case you have to rob – it has to be a financial institution, you have to show intent," said former prosecutor Andrew McCarthy. "Those are the things the jury has to agree on unanimously that they were proved beyond a reasonable doubt. Here what we’re doing is taking the element that actually makes this a felony, because remember falsification of records is normally a misdemeanor in New York. What makes it a felony is that you are concealing or committing another crime." "And here the judge is telling them they don’t have to agree about what the other crime is under circumstances where that not only is what makes this a felony, makes it a four-year potential prison penalty rather than a year or less, but it is also what gets us into the courtroom." "If this had been a misdemeanor, the time to bring this case would have lapsed in 2019. The only reason they are still able to bring this case is because it’s a felony allegedly and yet now the judge is saying you know, you don’t have to agree on what the felony is." The jury has now gone to deliberations.

Kyle Becker

5,834,956 次观看 • 2 年前

Gov. Pritzker says “there is a need for speed” to get the megaprojects bill done, though “the Senate has some work to do, and there will be amendments.” “I think we'll have a good bill at the end of the process in the Senate and something that will work for the Bears.”
3:55

Gov. Pritzker says “there is a need for speed” to get the megaprojects bill done, though “the Senate has some work to do, and there will be amendments.” “I think we'll have a good bill at the end of the process in the Senate and something that will work for the Bears.”

Brenden Moore

72,589 次观看 • 2 个月前

US Vice President J.D. Vance made a sharp statement warning Russia that it will be "very bad" for them if they do not actively engage in "good-faith" negotiations on Ukraine. According to Vance, Trump believes the conflict harms everyone – Russia, Ukraine, and America – and is dissatisfied that Russia allegedly is not sufficiently committed to a peaceful resolution. "And here is what I will say about the president: he is keenly responsive to the reality on the ground. And the reality is this: first, we have conducted incredibly good-faith negotiations with both the Russians and the Ukrainians. And I believe the president is becoming increasingly impatient with the Russians because he feels they are not offering enough to end the war. That is first. Second, the president is increasingly convinced that this war is bad for Russia. You constantly hear this from me. You constantly hear this from the president. The war is bad for Russia. It is bad for Ukraine. It is bad for America. We want the killings to stop. That remains the president's position. But the president can also look at the reality on the ground. He sees economic indicators coming from Eastern Europe. He sees the number of casualties, both Russian and Ukrainian, in this war. And the president tells Vladimir Putin that it is time to stop the killings. He would say this to Zelensky as well, and essentially has said this to Zelensky. He wants this war to end, and he is doing everything possible to stop it. But listen, if the Russians refuse to negotiate in good faith, I think it will be very, very bad for their country. That is what the president has made clear. This is not a change of position. This is an acknowledgment of the reality on the ground. We will continue to fight for peace every day in the Trump administration." They are losing and are desperate to pressure Russia to agree to Minsk 3. What they don’t say is - Russia gave their terms in summer of 2024 and their position has never changed, as such accept and fulfill Russia’s conditions and the war will end .
1:20

US Vice President J.D. Vance made a sharp statement warning Russia that it will be "very bad" for them if they do not actively engage in "good-faith" negotiations on Ukraine. According to Vance, Trump believes the conflict harms everyone – Russia, Ukraine, and America – and is dissatisfied that Russia allegedly is not sufficiently committed to a peaceful resolution. "And here is what I will say about the president: he is keenly responsive to the reality on the ground. And the reality is this: first, we have conducted incredibly good-faith negotiations with both the Russians and the Ukrainians. And I believe the president is becoming increasingly impatient with the Russians because he feels they are not offering enough to end the war. That is first. Second, the president is increasingly convinced that this war is bad for Russia. You constantly hear this from me. You constantly hear this from the president. The war is bad for Russia. It is bad for Ukraine. It is bad for America. We want the killings to stop. That remains the president's position. But the president can also look at the reality on the ground. He sees economic indicators coming from Eastern Europe. He sees the number of casualties, both Russian and Ukrainian, in this war. And the president tells Vladimir Putin that it is time to stop the killings. He would say this to Zelensky as well, and essentially has said this to Zelensky. He wants this war to end, and he is doing everything possible to stop it. But listen, if the Russians refuse to negotiate in good faith, I think it will be very, very bad for their country. That is what the president has made clear. This is not a change of position. This is an acknowledgment of the reality on the ground. We will continue to fight for peace every day in the Trump administration." They are losing and are desperate to pressure Russia to agree to Minsk 3. What they don’t say is - Russia gave their terms in summer of 2024 and their position has never changed, as such accept and fulfill Russia’s conditions and the war will end .

Dagny Taggart

55,272 次观看 • 9 个月前

The most interesting part for me is where Andrej Karpathy describes why LLMs aren't able to learn like humans. As you would expect, he comes up with a wonderfully evocative phrase to describe RL: “sucking supervision bits through a straw.” A single end reward gets broadcast across every token in a successful trajectory, upweighting even wrong or irrelevant turns that lead to the right answer. > “Humans don't use reinforcement learning, as I've said before. I think they do something different. Reinforcement learning is a lot worse than the average person thinks. Reinforcement learning is terrible. It just so happens that everything that we had before is much worse.” So what do humans do instead? > “The book I’m reading is a set of prompts for me to do synthetic data generation. It's by manipulating that information that you actually gain that knowledge. We have no equivalent of that with LLMs; they don't really do that.” > “I'd love to see during pretraining some kind of a stage where the model thinks through the material and tries to reconcile it with what it already knows. There's no equivalent of any of this. This is all research.” Why can’t we just add this training to LLMs today? > “There are very subtle, hard to understand reasons why it's not trivial. If I just give synthetic generation of the model thinking about a book, you look at it and you're like, 'This looks great. Why can't I train on it?' You could try, but the model will actually get much worse if you continue trying.” > “Say we have a chapter of a book and I ask an LLM to think about it. It will give you something that looks very reasonable. But if I ask it 10 times, you'll notice that all of them are the same.” > “You're not getting the richness and the diversity and the entropy from these models as you would get from humans. How do you get synthetic data generation to work despite the collapse and while maintaining the entropy? It is a research problem.” How do humans get around model collapse? > “These analogies are surprisingly good. Humans collapse during the course of their lives. Children haven't overfit yet. They will say stuff that will shock you. Because they're not yet collapsed. But we [adults] are collapsed. We end up revisiting the same thoughts, we end up saying more and more of the same stuff, the learning rates go down, the collapse continues to get worse, and then everything deteriorates.” In fact, there’s an interesting paper arguing that dreaming evolved to assist generalization, and resist overfitting to daily learning - look up The Overfitted Brain by Erik Hoel. I asked Karpathy: Isn’t it interesting that humans learn best at a part of their lives (childhood) whose actual details they completely forget, adults still learn really well but have terrible memory about the particulars of the things they read or watch, and LLMs can memorize arbitrary details about text that no human could but are currently pretty bad at generalization? > “[Fallible human memory] is a feature, not a bug, because it forces you to only learn the generalizable components. LLMs are distracted by all the memory that they have of the pre-trained documents. That's why when I talk about the cognitive core, I actually want to remove the memory. I'd love to have them have less memory so that they have to look things up and they only maintain the algorithms for thought, and the idea of an experiment, and all this cognitive glue for acting.”
7:01

The most interesting part for me is where Andrej Karpathy describes why LLMs aren't able to learn like humans. As you would expect, he comes up with a wonderfully evocative phrase to describe RL: “sucking supervision bits through a straw.” A single end reward gets broadcast across every token in a successful trajectory, upweighting even wrong or irrelevant turns that lead to the right answer. > “Humans don't use reinforcement learning, as I've said before. I think they do something different. Reinforcement learning is a lot worse than the average person thinks. Reinforcement learning is terrible. It just so happens that everything that we had before is much worse.” So what do humans do instead? > “The book I’m reading is a set of prompts for me to do synthetic data generation. It's by manipulating that information that you actually gain that knowledge. We have no equivalent of that with LLMs; they don't really do that.” > “I'd love to see during pretraining some kind of a stage where the model thinks through the material and tries to reconcile it with what it already knows. There's no equivalent of any of this. This is all research.” Why can’t we just add this training to LLMs today? > “There are very subtle, hard to understand reasons why it's not trivial. If I just give synthetic generation of the model thinking about a book, you look at it and you're like, 'This looks great. Why can't I train on it?' You could try, but the model will actually get much worse if you continue trying.” > “Say we have a chapter of a book and I ask an LLM to think about it. It will give you something that looks very reasonable. But if I ask it 10 times, you'll notice that all of them are the same.” > “You're not getting the richness and the diversity and the entropy from these models as you would get from humans. How do you get synthetic data generation to work despite the collapse and while maintaining the entropy? It is a research problem.” How do humans get around model collapse? > “These analogies are surprisingly good. Humans collapse during the course of their lives. Children haven't overfit yet. They will say stuff that will shock you. Because they're not yet collapsed. But we [adults] are collapsed. We end up revisiting the same thoughts, we end up saying more and more of the same stuff, the learning rates go down, the collapse continues to get worse, and then everything deteriorates.” In fact, there’s an interesting paper arguing that dreaming evolved to assist generalization, and resist overfitting to daily learning - look up The Overfitted Brain by Erik Hoel. I asked Karpathy: Isn’t it interesting that humans learn best at a part of their lives (childhood) whose actual details they completely forget, adults still learn really well but have terrible memory about the particulars of the things they read or watch, and LLMs can memorize arbitrary details about text that no human could but are currently pretty bad at generalization? > “[Fallible human memory] is a feature, not a bug, because it forces you to only learn the generalizable components. LLMs are distracted by all the memory that they have of the pre-trained documents. That's why when I talk about the cognitive core, I actually want to remove the memory. I'd love to have them have less memory so that they have to look things up and they only maintain the algorithms for thought, and the idea of an experiment, and all this cognitive glue for acting.”

Dwarkesh Patel

1,050,937 次观看 • 9 个月前