Video wird geladen...

Video konnte nicht geladen werden

Beim Laden dieses Videos ist ein Problem aufgetreten. Dies könnte an einem vorübergehenden Netzwerkproblem liegen oder das Video ist möglicherweise nicht verfügbar.

I've posed a question: "Does OpenAI have a better pre-train in the back pocket or not?" and looks like the answer from Mark Chen is "Yes" - so we should expect a big response from OpenAI before the end of the year. Mark Chen, OpenAI Head of Research: -... "In the last 2 years we've put so much resourcing into, into reasoning and one byproduct of that is you lose a little bit of muscle on pre training and post training." - "In the last six months, Jakub Pachocki and I have done a lot of work to build that muscle back up." - "With all the focus on RL, there's an alpha for us because we think there's so much room left in pre training." - "As a result of these efforts, we've been training much stronger models. And that also gives us a lot of confidence carrying into Gemini 3 and other releases coming this end of the year." So it does look like my assessment was broadly correct and they were neglecting pre-training quite a bit. The good news is that they've spent the last 6 months building that up and the big response shouldn't be too far away.show more

Peter Gostev

5,785 subscribers

84,628 Aufrufe • vor 8 Monaten •via X (Twitter)

Bildung Wissenschaft & Technologie

Anya Rossi• Live Now

Private livecam show

0 Kommentare

Keine Kommentare verfügbar

Kommentare vom Original-Post werden hier angezeigt

Ähnliche Videos

This week we’re getting GPT 5.5 and I couldn’t be more excited to see the first new pre train base in a while Before Spud releases, here’s the last video you need to watch - Greg Brockman perfectly indicated to us 3 important things that perfectly encapsulate the lineage of OpenAI and the mission. 1.This is the culmination of 2 years of research and is a fresh base pre-train that we haven’t gotten for a long time. 2.This is an “exciting release” and it will bring new capabilities (sort of a given but still very grateful nonetheless) & that it is AN EARLY VERSION of what we have coming - meaning they have a lot of RL to do to make this model even better since it’s a fresh pre-train. 3.THIS IS THE BEGINNING OF THE FLYWHEEL. As he said toward the end, “we have this engine of progress that just moves faster and faster and Spud is just one step along the way.” Depending on the capabilities we see this week from Spud, we may be able to map out if AGI will be sooner rather than later, judging by how much recursive work can exist within this model and the harness. In other words, the flywheel.

This week we’re getting GPT 5.5 and I couldn’t be more excited to see the first new pre train base in a while Before Spud releases, here’s the last video you need to watch - Greg Brockman perfectly indicated to us 3 important things that perfectly encapsulate the lineage of OpenAI and the mission. 1.This is the culmination of 2 years of research and is a fresh base pre-train that we haven’t gotten for a long time. 2.This is an “exciting release” and it will bring new capabilities (sort of a given but still very grateful nonetheless) & that it is AN EARLY VERSION of what we have coming - meaning they have a lot of RL to do to make this model even better since it’s a fresh pre-train. 3.THIS IS THE BEGINNING OF THE FLYWHEEL. As he said toward the end, “we have this engine of progress that just moves faster and faster and Spud is just one step along the way.” Depending on the capabilities we see this week from Spud, we may be able to map out if AGI will be sooner rather than later, judging by how much recursive work can exist within this model and the harness. In other words, the flywheel.

Chris

103,198 Aufrufe • vor 4 Monaten

US Secretary of State Marco Rubio: It is in my personal view that we've had probably the most productive and meaningful meeting so far in this entire process since we've been involved in it from the beginning. So I think the takeaway from it is I think this is a very, very meaningful, I would say probably best meeting and day we've had so far in this entire process going back to when we first came into office in January. We have a very good work product that was already built on a foundation of input from all the relative parties involved here, and we were able to go through some of those items now point by point, and I think we've made good progress. And then obviously there's the Russian side of the equation, but again, we think we have some pretty substantial insights over the last nine months into some of the things that are really important to them.

US Secretary of State Marco Rubio: It is in my personal view that we've had probably the most productive and meaningful meeting so far in this entire process since we've been involved in it from the beginning. So I think the takeaway from it is I think this is a very, very meaningful, I would say probably best meeting and day we've had so far in this entire process going back to when we first came into office in January. We have a very good work product that was already built on a foundation of input from all the relative parties involved here, and we were able to go through some of those items now point by point, and I think we've made good progress. And then obviously there's the Russian side of the equation, but again, we think we have some pretty substantial insights over the last nine months into some of the things that are really important to them.

Anton Gerashchenko

281,149 Aufrufe • vor 9 Monaten

“We have a crisis of open source models in the Western world. Outside of China, there are no good open source models. We don't even have any in the US now. The talent, the capital and the focus to be best-in-class at pre-training, mid-training, post-training is an extremely scarce skillset. The answer for a lot of countries may just be to take an open weights model from China, post-train or fine-tune their own version, and have that be what they start from." Everett Randle Why does the West have such poor open source models and is it a national security threat to rely on open-source Chinese models Demis Hassabis Yann LeCun Aidan Gomez Lilian Weng Jan Leike

“We have a crisis of open source models in the Western world. Outside of China, there are no good open source models. We don't even have any in the US now. The talent, the capital and the focus to be best-in-class at pre-training, mid-training, post-training is an extremely scarce skillset. The answer for a lot of countries may just be to take an open weights model from China, post-train or fine-tune their own version, and have that be what they start from." Everett Randle Why does the West have such poor open source models and is it a national security threat to rely on open-source Chinese models Demis Hassabis Yann LeCun Aidan Gomez Lilian Weng Jan Leike

Harry Stebbings

63,738 Aufrufe • vor 2 Monaten

Arteta: “We know what is at stake and the fact that, as a club, we haven't done it for so long is a reason for that. It talks about the difficulty but I just want to focus on the things that we can control, dominate, and we have done. “It's not that we have to do a lot of new things that we haven't done. No, no, no, let's keep doing a lot of it or almost all the things that we've done for nine months, because that's what has taken us all the way there. “On Wednesday, I saw a lot of them. And today we've gone back so we need to learn, we need to be very critical of ourselves, and then pick ourselves up because on Wednesday we have a big one coming up.”

Arteta: “We know what is at stake and the fact that, as a club, we haven't done it for so long is a reason for that. It talks about the difficulty but I just want to focus on the things that we can control, dominate, and we have done. “It's not that we have to do a lot of new things that we haven't done. No, no, no, let's keep doing a lot of it or almost all the things that we've done for nine months, because that's what has taken us all the way there. “On Wednesday, I saw a lot of them. And today we've gone back so we need to learn, we need to be very critical of ourselves, and then pick ourselves up because on Wednesday we have a big one coming up.”

Connor Humm

21,397 Aufrufe • vor 4 Monaten

"One of the very confusing things about the models right now: how to reconcile the fact that they are doing so well on evals. And you look at the evals and you go, 'Those are pretty hard evals.' But the economic impact seems to be dramatically behind. There is [a possible] explanation. Back when people were doing pre-training, the question of what data to train on was answered, because that answer was everything. So you don't have to think if it's going to be this data or that data. When people do RL training, they say, 'Okay, we want to have this kind of RL training for this thing and that kind of RL training for that thing.' You say, 'Hey, I would love our model to do really well when we release it. I want the evals to look great. What would be RL training that could help on this task?' If you combine this with generalization of the models actually being inadequate, that has the potential to explain a lot of what we are seeing, this disconnect between eval performance and actual real-world performance"

"One of the very confusing things about the models right now: how to reconcile the fact that they are doing so well on evals. And you look at the evals and you go, 'Those are pretty hard evals.' But the economic impact seems to be dramatically behind. There is [a possible] explanation. Back when people were doing pre-training, the question of what data to train on was answered, because that answer was everything. So you don't have to think if it's going to be this data or that data. When people do RL training, they say, 'Okay, we want to have this kind of RL training for this thing and that kind of RL training for that thing.' You say, 'Hey, I would love our model to do really well when we release it. I want the evals to look great. What would be RL training that could help on this task?' If you combine this with generalization of the models actually being inadequate, that has the potential to explain a lot of what we are seeing, this disconnect between eval performance and actual real-world performance"

Dwarkesh Patel

502,416 Aufrufe • vor 9 Monaten

#F1Testing || Lewis Hamilton was asked if this was the most optimistic he’s felt at this stage in the last 5 years. “Don’t remember the last 5 years but maybe yeah. I mean, I think I always try to enter into a season with confidence. But, of course, you’re face with all sorts of different challenges along the way. I think, this is definitely, for me, obviously, gone through quite a bit, and you know for me, left everything, all of last years behind me. I’ve spent a lot of time rebuilding this winter, refocusing, really getting into my body and my mind to a much better place. Obviously I’ve got a bit of a cold right now..” “But in general, just making sure that I’m able to arrive feeling better and honestly I genuinely feel, personally, in the best place that I’ve been in a long long time. With rearranging things within my team and then…” “Yeah for the car, I think we started off quite well so far. And it’s an exciting time with this new generation of car as well because it’s all brand new, we’re all trying to figure it out on the go. And as last year, we were locked into a car that, ultimately I inherited. This [SF-26] is a car that I’ve been able to be a part of developing on the simulator for the last 8-10 months and so like a bit of my DNA is within it so I’m more connected to this one for sure.”

sin ⁴⁴

89,926 Aufrufe • vor 6 Monaten

Ilya Sutskever says the AI scaling era is running out of road because pre-training is about to hit a finite-data wall. "At some point though, pre-training will run out of data. The data is very clearly finite." "Then, okay, what do you do next? Either you do some kind of a souped-up pre-training... or you're doing RL, or maybe something else." "In some sense, we are back to the age of research." His timeline is the part most AI investors and founders should hear: "From 2012 to 2020, it was the age of research. Now from 2020 to 2025, it was the age of scaling... But now the scale is so big." "Is the belief really that... if you had 100x more everything would be so different?" "I don't think that's true. So it's back to the age of research again, just with big computers."

Ilya Sutskever says the AI scaling era is running out of road because pre-training is about to hit a finite-data wall. "At some point though, pre-training will run out of data. The data is very clearly finite." "Then, okay, what do you do next? Either you do some kind of a souped-up pre-training... or you're doing RL, or maybe something else." "In some sense, we are back to the age of research." His timeline is the part most AI investors and founders should hear: "From 2012 to 2020, it was the age of research. Now from 2020 to 2025, it was the age of scaling... But now the scale is so big." "Is the belief really that... if you had 100x more everything would be so different?" "I don't think that's true. So it's back to the age of research again, just with big computers."

Karl Mehta

200,003 Aufrufe • vor 2 Monaten

McLaren's Chief Designer Rob Marshall explains the new MCL40 design: “It's all new car, front to back. New regulations. As people have touched upon, the car's a bit shorter. It's actually quite a lot shorter than previous cars, a lot lighter, coupled with a brand new power unit, 50-50 hybrid, and a new set of air regulations designed to try and reduce the size of the weight behind the car, make overtaking a bit easier. If you start from the front, the new front wing is still sort of arrowhead like the previous generation, but it's a bit lower and have a much broader and wider footplate, end plate, than we used to have. And also the front wing is now actuated much like the old DRS. These new cars have got a straight line mode where both the front and rear wings will move their flaps to reduce the drag on the car and help the car get down the straights faster. There's quite a lot of freedom in how you actuate that. I think we'll see different solutions from different cars on the grid. Moving a bit further back, you've got the crash structure, which is basically from here to here. That's all new again. The regulations have changed this year where we need to make sure that after a small shunt that's enough to knock the front wing off, maybe the front half of the nose, the remaining part still serves its function as saving the driver in a secondary crash against another barrier. So that's significantly complicated the design work going into the nose. And then we've got front suspension - again, last year's front suspension was I think very innovative and so is this. We've done quite a few changes which I hope benefit us. And as we go rearward and we've got the main body of the chassis. Again, all new regulations, much tougher homologation requirements. So the crash tests and the squeezes that go into the chassis are quite brutal this year. And a lot of effort and research has gone in to try and make the car able to withstand those. Also, as I touched upon, the car is so much shorter. So a lot of the packaging of where radiators and electrical boxes, which were typically scattered around the car, finding homes for those has been very difficult. There's just less space to put them all. What's helped us there is the fuel tank is a bit smaller. We're going to use less fuel in the race. From an aerodynamic point of view, in this region here, gone are the very curved underfloors that we had with last year's cars. We're replaced with a much more flat bottom car like we had in previous generations. And then as we go rearward again, our rear suspension is actually a bit of an iteration of last year's. We're very happy with what we did last year, and so we just built upon that. Redesigned the gearbox to make it lighter and shorter for the obvious reasons. And then the rear wing is similar-ish to last year's, and the actuation mechanism is a bit like old-school DRS, but it's now mounted on two pylons, and yeah, it will now operate in conjunction with the front wing, as I said.”

naenia ¹ ⁶³ | ia

97,794 Aufrufe • vor 6 Monaten

U7 Pre Training Conversation I like to hold a small interactive conversation post warm up, but pre primary focus. It helps to focus the players, and check for understanding . Also if you include some self reflection they have a tendency to take that self reflection into the training The 4 minutes spent in discussion is super valuable

U7 Pre Training Conversation I like to hold a small interactive conversation post warm up, but pre primary focus. It helps to focus the players, and check for understanding . Also if you include some self reflection they have a tendency to take that self reflection into the training The 4 minutes spent in discussion is super valuable

Michael Carroll 🇺🇸

17,562 Aufrufe • vor 2 Jahren

Anthropic's Sholto Douglas describes the concept of a software-only singularity: "It's one where the models are far better at digital tasks than they are at physical ones. And so we see rapid change in the digital world with models, and relatively little change in the physical world. So information and software changes dramatically, and this ends up having some pretty weird effects." "It means that maybe like the drivers of what have been the last couple decades of progress in the economy turn around." "And I think we'll see that flow on into the physical world but at a delay. So you get much better at doing chip design. You get much better at training AI models. AI models get a lot faster. Chips get a lot better. The general economy gets a lot more efficient because the sort of information and message parsing that is much of the rest of the economy ends up becoming much more efficient." "But at the same time you don't yet have robots providing limitless physical abundance. Science probably progresses really fast up to the degree that you need interaction with labs or larger particle colliders or something like this. And then you go, 'Okay well I need to build the robots.'"

Anthropic's Sholto Douglas describes the concept of a software-only singularity: "It's one where the models are far better at digital tasks than they are at physical ones. And so we see rapid change in the digital world with models, and relatively little change in the physical world. So information and software changes dramatically, and this ends up having some pretty weird effects." "It means that maybe like the drivers of what have been the last couple decades of progress in the economy turn around." "And I think we'll see that flow on into the physical world but at a delay. So you get much better at doing chip design. You get much better at training AI models. AI models get a lot faster. Chips get a lot better. The general economy gets a lot more efficient because the sort of information and message parsing that is much of the rest of the economy ends up becoming much more efficient." "But at the same time you don't yet have robots providing limitless physical abundance. Science probably progresses really fast up to the degree that you need interaction with labs or larger particle colliders or something like this. And then you go, 'Okay well I need to build the robots.'"

TBPN

208,090 Aufrufe • vor 6 Monaten

David Sacks: “FDA for AI” is fake news, but here’s why it’s making headlines @jason: “ Who's leading Trump down the path of regulation and creating this AI FDA?” David Sacks: “I think there's several things going on here. The first one is, there's a lot of fake news. This whole idea of an FDA for AI, I don't think any senior official supports it. Certainly, I don't think that's the way the president thinks about these issues. He's the most pro-innovation president we've ever had. And the White House Chief of Staff, Susie Wiles, just put out a statement last night that I think pretty much shoots this down. Second, there's another thing going on, which is a straw manning of what the Trump administration did on AI in its first year. In the same way that they want to spin this FDA for AI, they're also trying to spin what we did as this completely laissez-faire attitude, where there'd be no regulations whatsoever, nor guardrails. It's a way of criticizing what we did. They're trying to portray it as unsafe. In fact, if you look, on March 20th, the White House released a national AI regulatory framework in which we put out a four-page bulleted list of legislation that we would support. So we have not been against every conceivable regulation or every conceivable law, we just believe that there should be specific solutions to specific problems, as opposed to a giant power grab by Washington that would squash innovation. Point number three is, there is a legitimate thing happening here with, let's call it Mythos or cyber. Within 3-6 months, all the major frontier labs, including Chinese models, will have cyber capabilities. In response to that, we do need there to be a hardening of systems, and we do need there to be a scanning of codebases to find these vulnerabilities and patch them before the hackers do it. Because the hackers will have these capabilities in a matter of months. That's a certainty. So we do need a response to that. Now, my view on what should that response be, first of all, we should want the government and the private sector to work cooperatively, and I think they are. What we should be doing, I think, is getting these tools, Mythos, and then the OpenAI model, and others like it, in the hands of our cybersecurity industry. And by the way, not just the public companies like Palo Alto Networks and CrowdStrike, although certainly they're two of the most noteworthy, but there's also some incredibly strong startups on the way up. We need to get these tools into their hands as quickly as possible because they're a force multiplier for all the companies out there that aren't that good at cybersecurity, they can use these companies as vendors. And just one last point on this whole thing is, both Anthropic and OpenAI acted responsibly here. No one was trying to release these super powerful models. So in a way, all the people who are saying that we need pre-release approvals for models, they're trying to solve a problem that didn't exist. Yes, we do have this cyber issue, but that is a problem that we will solve over the next six months. What they're trying to do is use that issue to try and create a permanent new infrastructure in Washington. The classic 'never let a crisis go to waste' strategy.”

David Sacks: “FDA for AI” is fake news, but here’s why it’s making headlines @jason: “ Who's leading Trump down the path of regulation and creating this AI FDA?” David Sacks: “I think there's several things going on here. The first one is, there's a lot of fake news. This whole idea of an FDA for AI, I don't think any senior official supports it. Certainly, I don't think that's the way the president thinks about these issues. He's the most pro-innovation president we've ever had. And the White House Chief of Staff, Susie Wiles, just put out a statement last night that I think pretty much shoots this down. Second, there's another thing going on, which is a straw manning of what the Trump administration did on AI in its first year. In the same way that they want to spin this FDA for AI, they're also trying to spin what we did as this completely laissez-faire attitude, where there'd be no regulations whatsoever, nor guardrails. It's a way of criticizing what we did. They're trying to portray it as unsafe. In fact, if you look, on March 20th, the White House released a national AI regulatory framework in which we put out a four-page bulleted list of legislation that we would support. So we have not been against every conceivable regulation or every conceivable law, we just believe that there should be specific solutions to specific problems, as opposed to a giant power grab by Washington that would squash innovation. Point number three is, there is a legitimate thing happening here with, let's call it Mythos or cyber. Within 3-6 months, all the major frontier labs, including Chinese models, will have cyber capabilities. In response to that, we do need there to be a hardening of systems, and we do need there to be a scanning of codebases to find these vulnerabilities and patch them before the hackers do it. Because the hackers will have these capabilities in a matter of months. That's a certainty. So we do need a response to that. Now, my view on what should that response be, first of all, we should want the government and the private sector to work cooperatively, and I think they are. What we should be doing, I think, is getting these tools, Mythos, and then the OpenAI model, and others like it, in the hands of our cybersecurity industry. And by the way, not just the public companies like Palo Alto Networks and CrowdStrike, although certainly they're two of the most noteworthy, but there's also some incredibly strong startups on the way up. We need to get these tools into their hands as quickly as possible because they're a force multiplier for all the companies out there that aren't that good at cybersecurity, they can use these companies as vendors. And just one last point on this whole thing is, both Anthropic and OpenAI acted responsibly here. No one was trying to release these super powerful models. So in a way, all the people who are saying that we need pre-release approvals for models, they're trying to solve a problem that didn't exist. Yes, we do have this cyber issue, but that is a problem that we will solve over the next six months. What they're trying to do is use that issue to try and create a permanent new infrastructure in Washington. The classic 'never let a crisis go to waste' strategy.”

The All-In Podcast

150,106 Aufrufe • vor 3 Monaten

Cooper Flagg on the fan energy now compared to earlier in the season: “I think we have not given the fans a lot to love. So I think that it’s part of it, and I think we take a lot of responsibility on that and having lost the fans a little bit. And I think had we given them a lot more to cheer for and be up in the fourth quarters and be into it and giving them more wins, I think we might look at it a little differently, so, I think we can take a lot of responsibility for that, but the fans have been there, they stepped up tonight. They were huge, they were loud, they were positive, so I just want to look forward and try to continue to give them things to be positive for…I’m just glad they show up for us.”

Cooper Flagg on the fan energy now compared to earlier in the season: “I think we have not given the fans a lot to love. So I think that it’s part of it, and I think we take a lot of responsibility on that and having lost the fans a little bit. And I think had we given them a lot more to cheer for and be up in the fourth quarters and be into it and giving them more wins, I think we might look at it a little differently, so, I think we can take a lot of responsibility for that, but the fans have been there, they stepped up tonight. They were huge, they were loud, they were positive, so I just want to look forward and try to continue to give them things to be positive for…I’m just glad they show up for us.”

Noah Weber

67,240 Aufrufe • vor 9 Monaten

New Course: Post-training of LLMs Learn to post-train and customize an LLM in this short course, taught by Banghua Zhu, Assistant Professor at the University of Washington University of Washington, and co-founder of @NexusflowX. Training an LLM to follow instructions or answer questions has two key stages: pre-training and post-training. In pre-training, it learns to predict the next word or token from large amounts of unlabeled text. In post-training, it learns useful behaviors such as following instructions, tool use, and reasoning. Post-training transforms a general-purpose token predictor—trained on trillions of unlabeled text tokens—into an assistant that follows instructions and performs specific tasks. Because it is much cheaper than pre-training, it is practical for many more teams to incorporate post-training methods into their workflows than pre-training. In this course, you’ll learn three common post-training methods—Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Online Reinforcement Learning (RL)—and how to use each one effectively. With SFT, you train the model on pairs of input and ideal output responses. With DPO, you provide both a preferred (chosen) and a less preferred (rejected) response and train the model to favor the preferred output. With RL, the model generates an output, receives a reward score based on human or automated feedback, and updates the model to improve performance. You’ll learn the basic concepts, common use cases, and principles for curating high-quality data for effective training. Through hands-on labs, you’ll download a pre-trained model from Hugging Face and post-train it using SFT, DPO, and RL to see how each technique shapes model behavior. In detail, you’ll: - Understand what post-training is, when to use it, and how it differs from pre-training. - Build an SFT pipeline to turn a base model into an instruct model. - Explore how DPO reshapes behavior by minimizing contrastive loss—penalizing poor responses and reinforcing preferred ones. - Implement a DPO pipeline to change the identity of a chat assistant. - Learn online RL methods such as Proximal Policy Optimization (PPO) and Group Relative Policy Optimization (GRPO), and how to design reward functions. - Train a model with GRPO to improve its math capabilities using a verifiable reward. Post-training is one of the most rapidly developing areas of LLM training. Whether you’re building a high-accuracy context-specific assistant, fine-tuning a model's tone, or improving task-specific accuracy, this course will give you experience with the most important techniques shaping how LLMs are post-trained today. Please sign up here:

New Course: Post-training of LLMs Learn to post-train and customize an LLM in this short course, taught by Banghua Zhu, Assistant Professor at the University of Washington University of Washington, and co-founder of @NexusflowX. Training an LLM to follow instructions or answer questions has two key stages: pre-training and post-training. In pre-training, it learns to predict the next word or token from large amounts of unlabeled text. In post-training, it learns useful behaviors such as following instructions, tool use, and reasoning. Post-training transforms a general-purpose token predictor—trained on trillions of unlabeled text tokens—into an assistant that follows instructions and performs specific tasks. Because it is much cheaper than pre-training, it is practical for many more teams to incorporate post-training methods into their workflows than pre-training. In this course, you’ll learn three common post-training methods—Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Online Reinforcement Learning (RL)—and how to use each one effectively. With SFT, you train the model on pairs of input and ideal output responses. With DPO, you provide both a preferred (chosen) and a less preferred (rejected) response and train the model to favor the preferred output. With RL, the model generates an output, receives a reward score based on human or automated feedback, and updates the model to improve performance. You’ll learn the basic concepts, common use cases, and principles for curating high-quality data for effective training. Through hands-on labs, you’ll download a pre-trained model from Hugging Face and post-train it using SFT, DPO, and RL to see how each technique shapes model behavior. In detail, you’ll: - Understand what post-training is, when to use it, and how it differs from pre-training. - Build an SFT pipeline to turn a base model into an instruct model. - Explore how DPO reshapes behavior by minimizing contrastive loss—penalizing poor responses and reinforcing preferred ones. - Implement a DPO pipeline to change the identity of a chat assistant. - Learn online RL methods such as Proximal Policy Optimization (PPO) and Group Relative Policy Optimization (GRPO), and how to design reward functions. - Train a model with GRPO to improve its math capabilities using a verifiable reward. Post-training is one of the most rapidly developing areas of LLM training. Whether you’re building a high-accuracy context-specific assistant, fine-tuning a model's tone, or improving task-specific accuracy, this course will give you experience with the most important techniques shaping how LLMs are post-trained today. Please sign up here:

Andrew Ng

125,146 Aufrufe • vor 1 Jahr

I’ve always been so amazed when I see clips like this. They remind me of a real life Karate Kid. While I have trained when I was younger up until last year, I’ve always admired how strong the will of people can be. This woman enlisted in a training program at a Shaolin temple, this is as tough and as real as it gets, each crushing blow is felt, the intense training is what keeps them whole and unaffected. This is the kind of training that turns people into warriors. The age old traditional ways have molded some of the fiercest fighters of the world, could you withstand this training? I guess the better question would be, could you handle this kind of backbreaking training?

I’ve always been so amazed when I see clips like this. They remind me of a real life Karate Kid. While I have trained when I was younger up until last year, I’ve always admired how strong the will of people can be. This woman enlisted in a training program at a Shaolin temple, this is as tough and as real as it gets, each crushing blow is felt, the intense training is what keeps them whole and unaffected. This is the kind of training that turns people into warriors. The age old traditional ways have molded some of the fiercest fighters of the world, could you withstand this training? I guess the better question would be, could you handle this kind of backbreaking training?

SonnyBoy🇺🇸

57,203 Aufrufe • vor 3 Monaten

Anthropic CEO Dario Amodei: AI Progress isn’t magic, it’s just compute, data, and training. "All the cleverness, all the techniques, all of the “we need a new method,” doesn’t matter very much. There are only a few things that matter, and I listed seven of them. One is how much raw compute you have. Two is the quantity of data you have. Three is the quality and distribution of the data, meaning it needs to be broad. Four is how long you train for. Five is an objective function that can scale to the moon. Pre-training is one. Another is an RL-style objective: you have a goal and you go reach it. That includes more objective rewards (like math and coding), and more subjective rewards (like RL from human feedback, or higher-order versions of that). Six and seven are about normalization/conditioning: keeping numerical stability so the huge blob of compute flows cleanly instead of blowing up or getting weird." --- From 'Dwarkesh Patel' YT Channel (link in comment)

Anthropic CEO Dario Amodei: AI Progress isn’t magic, it’s just compute, data, and training. "All the cleverness, all the techniques, all of the “we need a new method,” doesn’t matter very much. There are only a few things that matter, and I listed seven of them. One is how much raw compute you have. Two is the quantity of data you have. Three is the quality and distribution of the data, meaning it needs to be broad. Four is how long you train for. Five is an objective function that can scale to the moon. Pre-training is one. Another is an RL-style objective: you have a goal and you go reach it. That includes more objective rewards (like math and coding), and more subjective rewards (like RL from human feedback, or higher-order versions of that). Six and seven are about normalization/conditioning: keeping numerical stability so the huge blob of compute flows cleanly instead of blowing up or getting weird." --- From 'Dwarkesh Patel' YT Channel (link in comment)

Rohan Paul

75,803 Aufrufe • vor 6 Monaten

This last year and a half has been quite a challenging and equally awesome time. I’ve faced a lot of backlash for what I sing about, which in some ways has set me back, but in many other ways has opened incredible new doors. I had no idea there were so many of us who felt the same way. We’re fed up with much of what’s going on in the world. The systems that don’t serve us but pretend to. The institutions that rob us and keep us dependent. But what has been so beautiful is discovering how many of us feel this together and genuinely want to create a future that uplifts and serves humanity, and each other. There are so many people working for good, working for the light. I hope the music and art I make can contribute to that in a great way. So from the bottom of my heart, thank you. Thank you for having my back. I have yours too.

This last year and a half has been quite a challenging and equally awesome time. I’ve faced a lot of backlash for what I sing about, which in some ways has set me back, but in many other ways has opened incredible new doors. I had no idea there were so many of us who felt the same way. We’re fed up with much of what’s going on in the world. The systems that don’t serve us but pretend to. The institutions that rob us and keep us dependent. But what has been so beautiful is discovering how many of us feel this together and genuinely want to create a future that uplifts and serves humanity, and each other. There are so many people working for good, working for the light. I hope the music and art I make can contribute to that in a great way. So from the bottom of my heart, thank you. Thank you for having my back. I have yours too.

iyah may

27,747 Aufrufe • vor 28 Tagen

.David Sacks: "You're gonna see in the wake of this is that a lot of Jewish people are realizing that they don't have a home on the left anymore. And I expect that many Jews are going to start shifting right into the Republican Party, to a place where I've been for a while.... But I think what's happened over the last few decades is that the civil rights movement, in particular, and the left have moved to this woke ideology where it's no longer about colorblindness. It's more about identity groups. And instead of trying to get past racial differences, it's been about accentuating them. And so we've had this whole equity agenda, which is really defined as redistribution from one racial group to another racial group. I think that, for whatever reason, a lot of Jews just hadn't confronted the reality that the left had really changed in this way. I think it goes back to the fact that they thought if we're going to be defining identity groups in this woke way, Jews obviously should be one of these victim groups, but they're waking up to the fact that Jews are not. Jews are in the minds of this woke ideology, just white people with a Jewish background. As a result, they're part of an oppressor class. And I think that a lot of Jewish people who are waking up to this are realizing this is actually a very destructive ideology, and it makes us the bad guys. So I would expect that, again, a lot of Jewish people are waking up to how the left has changed... and I would expect there to be kind of a pilgrimage now of more Jews in America towards the right, as opposed to remaining on the left where they've always been." Via The All-In Podcast:

.David Sacks: "You're gonna see in the wake of this is that a lot of Jewish people are realizing that they don't have a home on the left anymore. And I expect that many Jews are going to start shifting right into the Republican Party, to a place where I've been for a while.... But I think what's happened over the last few decades is that the civil rights movement, in particular, and the left have moved to this woke ideology where it's no longer about colorblindness. It's more about identity groups. And instead of trying to get past racial differences, it's been about accentuating them. And so we've had this whole equity agenda, which is really defined as redistribution from one racial group to another racial group. I think that, for whatever reason, a lot of Jews just hadn't confronted the reality that the left had really changed in this way. I think it goes back to the fact that they thought if we're going to be defining identity groups in this woke way, Jews obviously should be one of these victim groups, but they're waking up to the fact that Jews are not. Jews are in the minds of this woke ideology, just white people with a Jewish background. As a result, they're part of an oppressor class. And I think that a lot of Jewish people who are waking up to this are realizing this is actually a very destructive ideology, and it makes us the bad guys. So I would expect that, again, a lot of Jewish people are waking up to how the left has changed... and I would expect there to be kind of a pilgrimage now of more Jews in America towards the right, as opposed to remaining on the left where they've always been." Via The All-In Podcast:

KanekoaTheGreat

543,004 Aufrufe • vor 2 Jahren