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。

What happens when you change the hand mid-task? We tested this by modifying the hands mid-rollout and letting the same model keep running. It perceives the new tool, and finds a new trajectory and contact strategy to complete the task. This works because training on mixed data forces the... show more

Generalist

11,729 subscribers

21,858 次观看 • 27 天前 •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

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.

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

429,105 次观看 • 2 个月前

The renewal of fighting in Iran demonstrates the problem with Trump’s lack of strategy, the vagueness of the deal to make a deal, and the belligerence of an Iranian regime that feels it has the upper hand. This week, I’ll introduce a new War Powers Resolution to force the Senate to vote on Trump’s war and the price spikes it is causing.

The renewal of fighting in Iran demonstrates the problem with Trump’s lack of strategy, the vagueness of the deal to make a deal, and the belligerence of an Iranian regime that feels it has the upper hand. This week, I’ll introduce a new War Powers Resolution to force the Senate to vote on Trump’s war and the price spikes it is causing.

Senator Adam Schiff

50,801 次观看 • 1 个月前

Blendshape Viewer has been updated to V3. This tool lets you visualize which area of a model is affected by a blendshape, and change its value. From now on, you can press the ALT key to compare the blendshape. This version should also run faster on models with lots of shapes.

Blendshape Viewer has been updated to V3. This tool lets you visualize which area of a model is affected by a blendshape, and change its value. From now on, you can press the ALT key to compare the blendshape. This version should also run faster on models with lots of shapes.

Haï~

21,715 次观看 • 28 天前

President Zelenskyy: The task of Ukrainian units is to inflict a level of losses on the occupier at which Russian casualties exceed the amount of reinforcements they are able to send to their forces each month. This is a realistic task. When we speak of 50,000 Russian losses per month, that is the optimal level. It is a difficult task, without question, but it is precisely the level needed for Russia to begin weighing what it is doing and what it is fighting for. The task of the Ministry of Defense of Ukraine, our army, and all the Defense and Security Forces of Ukraine is to ensure exactly this level of Russian losses.

President Zelenskyy: The task of Ukrainian units is to inflict a level of losses on the occupier at which Russian casualties exceed the amount of reinforcements they are able to send to their forces each month. This is a realistic task. When we speak of 50,000 Russian losses per month, that is the optimal level. It is a difficult task, without question, but it is precisely the level needed for Russia to begin weighing what it is doing and what it is fighting for. The task of the Ministry of Defense of Ukraine, our army, and all the Defense and Security Forces of Ukraine is to ensure exactly this level of Russian losses.

Anton Gerashchenko

77,058 次观看 • 6 个月前

watch this anon. i gave NVIDIA's biggest model ever a single task. 100 minutes and 440,000 tokens later, it had rendered nothing. not one important thing on the screen. this is Nemotron 3 Ultra. 550 billion parameters, a hybrid Mamba Transformer MoE, the largest model NVIDIA has ever shipped, and they built it specifically for long-running agentic coding. so i handed it exactly that: build a 3D scene from a spec, multiple files, iterate until the tests pass. the same task a frontier model one shotted in minutes. i genuinely wanted to be impressed. it ran for an hour and forty. burned through 440,000 tokens. wrote every file, passed its own tests, and proudly printed "task complete."the browser was blank. the 3D scene never rendered. not once. and the long horizon agentic behavior was genuinely good. it stayed on task the whole hour and forty, wrote real multi-file code, drove its own tools without derailing. it just couldn't turn any of that into something that actually runs. here's the part that gets me. it's a text model, it cannot see its own output. so it sat there looping on a broken vision tool, trying to "look" at the page, hitting error after error, never once reasoning its way out. it declared victory on an empty screen because it had no way to know the screen was empty. to be fair, i genuinely don't know what quant the NIM was serving, so maybe some of that's on the serving, not the model. but the biggest model NVIDIA has ever made, on the exact task it was designed for, couldn't tell it had built nothing in 100 minutes. same task on a local model, below thread👇.

watch this anon. i gave NVIDIA's biggest model ever a single task. 100 minutes and 440,000 tokens later, it had rendered nothing. not one important thing on the screen. this is Nemotron 3 Ultra. 550 billion parameters, a hybrid Mamba Transformer MoE, the largest model NVIDIA has ever shipped, and they built it specifically for long-running agentic coding. so i handed it exactly that: build a 3D scene from a spec, multiple files, iterate until the tests pass. the same task a frontier model one shotted in minutes. i genuinely wanted to be impressed. it ran for an hour and forty. burned through 440,000 tokens. wrote every file, passed its own tests, and proudly printed "task complete."the browser was blank. the 3D scene never rendered. not once. and the long horizon agentic behavior was genuinely good. it stayed on task the whole hour and forty, wrote real multi-file code, drove its own tools without derailing. it just couldn't turn any of that into something that actually runs. here's the part that gets me. it's a text model, it cannot see its own output. so it sat there looping on a broken vision tool, trying to "look" at the page, hitting error after error, never once reasoning its way out. it declared victory on an empty screen because it had no way to know the screen was empty. to be fair, i genuinely don't know what quant the NIM was serving, so maybe some of that's on the serving, not the model. but the biggest model NVIDIA has ever made, on the exact task it was designed for, couldn't tell it had built nothing in 100 minutes. same task on a local model, below thread👇.

Sudo su

32,589 次观看 • 1 个月前

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 次观看 • 1 年前

🎙️What’s changing with the Championship Factor, and what does it mean for your school? At its June meeting, the MSHSAA Board of Directors adopted a new Championship Factor model. In our latest episode, MSHSAA leadership explains what led to the change, how the new model works, and what schools need to know moving forward. Listen here:

🎙️What’s changing with the Championship Factor, and what does it mean for your school? At its June meeting, the MSHSAA Board of Directors adopted a new Championship Factor model. In our latest episode, MSHSAA leadership explains what led to the change, how the new model works, and what schools need to know moving forward. Listen here:

MSHSAA

14,354 次观看 • 5 天前

A wild chimpanzee displayed a unique and touching behavior. He borrowed a photographer's hand to collect water from a puddle and drank it calmly. Afterward, the chimpanzee returned the favor by cleaning the man's hands with the same water, as if understanding and appreciating the assistance. This rare moment demonstrates the chimpanzee's remarkable intelligence, empathy, and social behavior.

A wild chimpanzee displayed a unique and touching behavior. He borrowed a photographer's hand to collect water from a puddle and drank it calmly. Afterward, the chimpanzee returned the favor by cleaning the man's hands with the same water, as if understanding and appreciating the assistance. This rare moment demonstrates the chimpanzee's remarkable intelligence, empathy, and social behavior.

ᑭEGEᒪ ᗰᒪᗩᖇᗩT

140,644 次观看 • 7 个月前

15–30 minutes of real-world robot data. That's now enough to go from sim-to-real failure to working robot. Let’s see… You train a robot in simulation. You deploy it in the real world. It fails. The physics don't match. So you try to fine-tune it with real data, but… you never have enough real data, and the fine-tuning breaks everything the simulation taught it. SimDist fixes this with one key decision: don't transfer the policy. Transfer the world model. Keep the reward and value knowledge from simulation frozen. Only update the part that's actually wrong, how the robot predicts physics. Now the robot doesn't have to relearn the entire task in the real world. It already knows what success looks like. It just needs to correct its understanding of how the real world moves. The part that makes this work: they also trained on failures and recoveries; not just perfect demonstrations. Without that, the planner finds the gaps and exploits them. With it, the robot can tell a good future from a bad one. That's all it needs. Results on peg insertion, table leg assembly, locomotion on slippery and uneven surfaces. Tasks that require precision, force, and quick reaction. Thanks for sharing, Tyler Westenbroek ([Tyler Westenbroek]. Interactive visualization + paper: ——- Weekly robotics and AI insights. Subscribe free:

15–30 minutes of real-world robot data. That's now enough to go from sim-to-real failure to working robot. Let’s see… You train a robot in simulation. You deploy it in the real world. It fails. The physics don't match. So you try to fine-tune it with real data, but… you never have enough real data, and the fine-tuning breaks everything the simulation taught it. SimDist fixes this with one key decision: don't transfer the policy. Transfer the world model. Keep the reward and value knowledge from simulation frozen. Only update the part that's actually wrong, how the robot predicts physics. Now the robot doesn't have to relearn the entire task in the real world. It already knows what success looks like. It just needs to correct its understanding of how the real world moves. The part that makes this work: they also trained on failures and recoveries; not just perfect demonstrations. Without that, the planner finds the gaps and exploits them. With it, the robot can tell a good future from a bad one. That's all it needs. Results on peg insertion, table leg assembly, locomotion on slippery and uneven surfaces. Tasks that require precision, force, and quick reaction. Thanks for sharing, Tyler Westenbroek ([Tyler Westenbroek]. Interactive visualization + paper: ——- Weekly robotics and AI insights. Subscribe free:

Ilir Aliu

14,214 次观看 • 3 个月前

How dangerous is Mercedes' strategy? Mercedes new strategy could dramatically change not just Germany's economy, but Europe's as a whole. After all, the automotive industry is the key economic sector. Mercedes is largely pulling out of the lower segments of small and cheaper cars to refocus on the luxury market. But will this strategy turn out to be successful? And what does it mean for you? What can you learn from it? Let's dig into these questions together and find the answers. More on this in the video. #mercedes #ElectricVehicles #automobile #foresight

How dangerous is Mercedes' strategy? Mercedes new strategy could dramatically change not just Germany's economy, but Europe's as a whole. After all, the automotive industry is the key economic sector. Mercedes is largely pulling out of the lower segments of small and cheaper cars to refocus on the luxury market. But will this strategy turn out to be successful? And what does it mean for you? What can you learn from it? Let's dig into these questions together and find the answers. More on this in the video. #mercedes #ElectricVehicles #automobile #foresight

Dr Pero Micic

23,542,463 次观看 • 2 年前

Data for the Real World Diana & Austin Tindle AI has become superhuman at code, language, and images because it has enormous amounts of data to learn from. But our understanding of the physical world is still limited by sparse sensors and models built largely on intuition. Cheaper sensors and better foundation models could unlock entirely new datasets for energy, agriculture, logistics, construction, and the climate. Once you can accurately model the physical world, you can begin to control it.

Data for the Real World Diana & Austin Tindle AI has become superhuman at code, language, and images because it has enormous amounts of data to learn from. But our understanding of the physical world is still limited by sparse sensors and models built largely on intuition. Cheaper sensors and better foundation models could unlock entirely new datasets for energy, agriculture, logistics, construction, and the climate. Once you can accurately model the physical world, you can begin to control it.

Y Combinator

22,185 次观看 • 28 天前

“You take a a Rolex Daytona today, and you look back at the first one in 1963. To me, the one in 1963 is better aesthetically. Because its the purest form of the concept. The purest form of the idea. Everything else from there has been layered on because they need to sell more and they need to come up with a new version and new model. This is what happens. They couldn't still sell that first version. That's not the way the world works. You have to keep making new stuff. The pure idea, the pure form of the thing [is best]”

“You take a a Rolex Daytona today, and you look back at the first one in 1963. To me, the one in 1963 is better aesthetically. Because its the purest form of the concept. The purest form of the idea. Everything else from there has been layered on because they need to sell more and they need to come up with a new version and new model. This is what happens. They couldn't still sell that first version. That's not the way the world works. You have to keep making new stuff. The pure idea, the pure form of the thing [is best]”

David Senra

105,087 次观看 • 5 个月前

"The death of the old model of the world is shocking and traumatic." That's Obama's former political director, in Toronto this weekend, admitting the empire's model is dead. @SJKokinda on what they're trying to replace it with — and why it won't work.

"The death of the old model of the world is shocking and traumatic." That's Obama's former political director, in Toronto this weekend, admitting the empire's model is dead. @SJKokinda on what they're trying to replace it with — and why it won't work.

Promethean Action

350,348 次观看 • 3 个月前

I cant believe this guy just made a permanent solution to context bloat and open sourced it all! when we tested this tool (Context+) for solving an issue on the OpenCode repository, the agent using this tool used ~6.5k fewer tokens, found the code and fixed it in half the time! the results were surprising: 6 to 10k tokens saved per prompt, completed task in ~2 minutes while the agent running without the tool took ~4 mins for the same and got stuck in loops bro built an entire beast by using all the modern tools that we could think of: undo trees, semantic search by meaning (by haskellforall), advanced refactoring, blast radius, advanced file context trees, restore points... i can keep going on semantic code search and context trees are the future of agentic coding and this tool proves it the feature i loved the most is semantic search and how it gets things done 2x faster with least possible tokens it makes an agent that actually knows what it’s doing and not just guessing, it makes meaning from your code similar to RAG. if you aren't optimizing your context, you are just burning money the developer says this tool is still under development, it can have unexpected behavior and the docs need updates but the video shows the reality of how fast it can be github: get here:

I cant believe this guy just made a permanent solution to context bloat and open sourced it all! when we tested this tool (Context+) for solving an issue on the OpenCode repository, the agent using this tool used ~6.5k fewer tokens, found the code and fixed it in half the time! the results were surprising: 6 to 10k tokens saved per prompt, completed task in ~2 minutes while the agent running without the tool took ~4 mins for the same and got stuck in loops bro built an entire beast by using all the modern tools that we could think of: undo trees, semantic search by meaning (by haskellforall), advanced refactoring, blast radius, advanced file context trees, restore points... i can keep going on semantic code search and context trees are the future of agentic coding and this tool proves it the feature i loved the most is semantic search and how it gets things done 2x faster with least possible tokens it makes an agent that actually knows what it’s doing and not just guessing, it makes meaning from your code similar to RAG. if you aren't optimizing your context, you are just burning money the developer says this tool is still under development, it can have unexpected behavior and the docs need updates but the video shows the reality of how fast it can be github: get here:

forloop

226,271 次观看 • 5 个月前

Karpathy said something you'll regret ignoring: "We have to keep the AI on the leash. I'm still the bottleneck. I have to make sure this thing isn't introducing bugs and that there's no security issues." He said it at YC talk last year, when the worry was reliability. The models hallucinated and made mistakes no human would, so the leash implied keeping yourself in the loop and checking the output before trusting it. The models are far better now, and the line still holds, for a reason he was not focused on back then. Even a model that writes flawless code today still has no idea who is allowed to run it. Correctness and authorization are different problems, and only correctness improves as the model improves. A perfect agent still hands a tool where anyone can do anything, because permission was never part of the task. I actually tested this in practice with Claude Code. I asked it to build a small internal tool with a button that issues account credits. It worked first try, and running it locally, the credit applied the instant I clicked. Nothing decided who was allowed to click it. The agent wrote the right logic and displayed a success notification. It never checked whether the caller had the right, whether it should pause for a human, or whether anything was logged. And this is not a bug a smarter model can outgrow because the leash was never in the code. Identity, permissions, and audit live in the system that runs the app, not in what the agent generates. To solve this, I took the exact same bundle and hosted it on Retool. The credit write that fired silently on my laptop now stopped at an approval gate, resolved to a real identity through SSO, and landed in an audit log. I wrote none of it. The app inherited the entire boundary the moment it was deployed, and the video shows the before and after. You can try it yourself here: I also wrote a detailed breakdown of the whole thing in my recent article, and I worked with the team to put this together. It walks through the build, the exact moment the credit write went through on my laptop with nobody checking, and then what changed when the same app ran on Retool. It also covers why this is a property of the runtime and not something a better model fixes, which is why devs typically miss this. The article is quoted below.

Karpathy said something you'll regret ignoring: "We have to keep the AI on the leash. I'm still the bottleneck. I have to make sure this thing isn't introducing bugs and that there's no security issues." He said it at YC talk last year, when the worry was reliability. The models hallucinated and made mistakes no human would, so the leash implied keeping yourself in the loop and checking the output before trusting it. The models are far better now, and the line still holds, for a reason he was not focused on back then. Even a model that writes flawless code today still has no idea who is allowed to run it. Correctness and authorization are different problems, and only correctness improves as the model improves. A perfect agent still hands a tool where anyone can do anything, because permission was never part of the task. I actually tested this in practice with Claude Code. I asked it to build a small internal tool with a button that issues account credits. It worked first try, and running it locally, the credit applied the instant I clicked. Nothing decided who was allowed to click it. The agent wrote the right logic and displayed a success notification. It never checked whether the caller had the right, whether it should pause for a human, or whether anything was logged. And this is not a bug a smarter model can outgrow because the leash was never in the code. Identity, permissions, and audit live in the system that runs the app, not in what the agent generates. To solve this, I took the exact same bundle and hosted it on Retool. The credit write that fired silently on my laptop now stopped at an approval gate, resolved to a real identity through SSO, and landed in an audit log. I wrote none of it. The app inherited the entire boundary the moment it was deployed, and the video shows the before and after. You can try it yourself here: I also wrote a detailed breakdown of the whole thing in my recent article, and I worked with the team to put this together. It walks through the build, the exact moment the credit write went through on my laptop with nobody checking, and then what changed when the same app ran on Retool. It also covers why this is a property of the runtime and not something a better model fixes, which is why devs typically miss this. The article is quoted below.

Akshay 🚀

42,911 次观看 • 1 个月前

Today I called for a hearing on the Four Pillars strategy to address public drug use, addiction, and overdoses in San Francisco Prevention, harm reduction, enforcement, and treatment. Zurich deployed this model and now it is one of the safest cities *in the world.*

Today I called for a hearing on the Four Pillars strategy to address public drug use, addiction, and overdoses in San Francisco Prevention, harm reduction, enforcement, and treatment. Zurich deployed this model and now it is one of the safest cities in the world.

Jackie Fielder

184,177 次观看 • 1 年前

DELETE FIGMA. ONE MCP CONNECTOR TURNS CLAUDE CODE INTO A GOD-MODE STUDIO IN 30 MINUTES. The exact setup, step by step, no jokes: > open connectors, add a custom connector, drop in the Higgsfield MCP > open a Claude Code session and paste one prompt > it generates the first frame, the last frame, and the motion between them > then wires the result straight into an animated hero section for your site > 30 minutes end to end, nothing rendered by hand The bigger thing hiding here: MCP means the model rents hands. it doesn't know how to animate, so it hires a tool that does, mid-task, without you switching tabs. The ceiling of what one prompt ships isn't the model anymore, it's how many hands you've plugged into it. The method underneath (the goal, the rules, the bar, the loop) is in the article below, and it works with any connector you add next.

DELETE FIGMA. ONE MCP CONNECTOR TURNS CLAUDE CODE INTO A GOD-MODE STUDIO IN 30 MINUTES. The exact setup, step by step, no jokes: > open connectors, add a custom connector, drop in the Higgsfield MCP > open a Claude Code session and paste one prompt > it generates the first frame, the last frame, and the motion between them > then wires the result straight into an animated hero section for your site > 30 minutes end to end, nothing rendered by hand The bigger thing hiding here: MCP means the model rents hands. it doesn't know how to animate, so it hires a tool that does, mid-task, without you switching tabs. The ceiling of what one prompt ships isn't the model anymore, it's how many hands you've plugged into it. The method underneath (the goal, the rules, the bar, the loop) is in the article below, and it works with any connector you add next.

Miraqle

38,373 次观看 • 1 个月前

"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

502,249 次观看 • 8 个月前