Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

.Polymath is training world generation models to automate the creation of RL environments. Traditionally, RL environment generation has been bottlenecked by human data. Superintelligence will never be achieved by human data alone. Polymath is building the core technology to enable automated environment generation using far less human effort than...

ycombinator's profile picture

Y Combinator

1,558,364 subscribers

44,192 Aufrufe • vor 7 Monaten •via X (Twitter)

BildungNachrichten & PolitikWissenschaft & Technologie
LIVE

Anya Rossi• Live Now

Private livecam show

0 Kommentare

Keine Kommentare verfügbar

Kommentare vom Original-Post werden hier angezeigt

Ähnliche Videos

Introducing arkenOS — an autonomous training gym that takes any RL agent from a goal to a proven, deployable policy. We are automating complex RL research, enabling high-velocity training and rigorous, diverse environment design and testing right on your local hardware. Out of the box. Our bet is simple: The future of AI will be built by companies giving millions of agents millions of simulated environments to adapt and learn cause and effect in, and survive. This continuous, autonomous selection will create a digital Cambrian explosion, distilling evolutionary trial and error into a foundational specialized intuition.
0:39

Introducing arkenOS — an autonomous training gym that takes any RL agent from a goal to a proven, deployable policy. We are automating complex RL research, enabling high-velocity training and rigorous, diverse environment design and testing right on your local hardware. Out of the box. Our bet is simple: The future of AI will be built by companies giving millions of agents millions of simulated environments to adapt and learn cause and effect in, and survive. This continuous, autonomous selection will create a digital Cambrian explosion, distilling evolutionary trial and error into a foundational specialized intuition.

Dev Shah

17,829 Aufrufe • vor 2 Monaten

A model is only as good as its data, and we’ve long since exhausted the internet. From here on out, model progress is gated by data production. ’s Brendan (can/do) joined us at our Sovereign AI event to talk about how RL environments get built, and why your data might be your real moat: 00:00 Introduction 00:47 A short history of the data market: crowdsourcing to agentic data 02:29 What an RL environment is: worlds, apps, tasks 03:57 Why only humans can measure the frontier 05:35 Building verifiers is the hard part 06:44 Walkthrough: a real legal RL environment 08:18 Leaderboards — and what open weights change 09:45 Post-training results on Apex Agents 11:17 Three ways companies buy data 12:49 Q&A: How do you price data? 14:17 Q&A: What "data quality" actually means 16:42 Q&A: The misunderstanding about synthetic data 18:17 Q&A: Why RL environments now — and what comes after 21:20 Q&A: Can you scale rubric generation with models? 23:00 Q&A: RL environments for cyber defense 25:33 Q&A: Build data in-house or partner?
26:21

A model is only as good as its data, and we’ve long since exhausted the internet. From here on out, model progress is gated by data production. ’s Brendan (can/do) joined us at our Sovereign AI event to talk about how RL environments get built, and why your data might be your real moat: 00:00 Introduction 00:47 A short history of the data market: crowdsourcing to agentic data 02:29 What an RL environment is: worlds, apps, tasks 03:57 Why only humans can measure the frontier 05:35 Building verifiers is the hard part 06:44 Walkthrough: a real legal RL environment 08:18 Leaderboards — and what open weights change 09:45 Post-training results on Apex Agents 11:17 Three ways companies buy data 12:49 Q&A: How do you price data? 14:17 Q&A: What "data quality" actually means 16:42 Q&A: The misunderstanding about synthetic data 18:17 Q&A: Why RL environments now — and what comes after 21:20 Q&A: Can you scale rubric generation with models? 23:00 Q&A: RL environments for cyber defense 25:33 Q&A: Build data in-house or partner?

Pat Grady

66,691 Aufrufe • vor 1 Monat

So I heard we need more data for robot learning :) Purely real world teleop is expensive and slow, making large scale data collection challenging. I’ve been excited about getting more data into robot learning, going beyond just real-world teleop data. To this end, we’ve been scaling up data generation with RL in realistic simulations generated on the fly from crowdsourced videos. Enables realistic data collection, much more cheaply than purely real world teleop. Importantly, data collection becomes even*cheaper* with more environments, allowing training with over 100x more data. Transfers to real robots for generalizable manipulation. A 🧵 (1/N)
0:41

So I heard we need more data for robot learning :) Purely real world teleop is expensive and slow, making large scale data collection challenging. I’ve been excited about getting more data into robot learning, going beyond just real-world teleop data. To this end, we’ve been scaling up data generation with RL in realistic simulations generated on the fly from crowdsourced videos. Enables realistic data collection, much more cheaply than purely real world teleop. Importantly, data collection becomes even*cheaper* with more environments, allowing training with over 100x more data. Transfers to real robots for generalizable manipulation. A 🧵 (1/N)

Abhishek Gupta

13,358 Aufrufe • vor 1 Jahr

Today I’m excited to introduce micro1 Intelligence, the world’s most advanced platform for training frontier AI models. Achieving AGI is bottlenecked by one main thing: high-quality data. Data based on real-world environments that capture human expert workflows, complex decision-making, and reward signals models need to learn. With micro1 Intelligence, frontier labs can train on RL environments across every subject matter all in one place.
1:33

Today I’m excited to introduce micro1 Intelligence, the world’s most advanced platform for training frontier AI models. Achieving AGI is bottlenecked by one main thing: high-quality data. Data based on real-world environments that capture human expert workflows, complex decision-making, and reward signals models need to learn. With micro1 Intelligence, frontier labs can train on RL environments across every subject matter all in one place.

Ali Ansari

333,257 Aufrufe • vor 10 Monaten

Open-ended coding training data may no longer be the bottleneck: AI can scale open-ended tasks—and even outperform human-expert curation. FrontierCS team is releasing FrontierSmith: a system for synthesizing open-ended coding problems at scale. Starting from closed-ended coding tasks, FrontierSmith mutates, filters, and builds runnable optimization environments for long-horizon coding agents. In our experiments, FrontierSmith data trains stronger models than human-curated open-ended data on FrontierCS and ALE-bench. Blog: Paper: Code: Model:
0:56

Open-ended coding training data may no longer be the bottleneck: AI can scale open-ended tasks—and even outperform human-expert curation. FrontierCS team is releasing FrontierSmith: a system for synthesizing open-ended coding problems at scale. Starting from closed-ended coding tasks, FrontierSmith mutates, filters, and builds runnable optimization environments for long-horizon coding agents. In our experiments, FrontierSmith data trains stronger models than human-curated open-ended data on FrontierCS and ALE-bench. Blog: Paper: Code: Model:

Qiuyang Mang

121,283 Aufrufe • vor 4 Monaten

Hiring RL Engineer! Started off as a curious project at Lossfunk to push the boundaries of LLMs in social reasoning - we are now building RL environments, data, and benchmarks to simulate more real-world scenarios. If you want to train SoTA RL models over multi-GPUs (H200s/B200s) to unlock next AI frontier, this is for you.
1:21

Hiring RL Engineer! Started off as a curious project at Lossfunk to push the boundaries of LLMs in social reasoning - we are now building RL environments, data, and benchmarks to simulate more real-world scenarios. If you want to train SoTA RL models over multi-GPUs (H200s/B200s) to unlock next AI frontier, this is for you.

Satpal Singh Rathore

46,010 Aufrufe • vor 1 Jahr

Today's Training Data episode takes us BTS on the infrastructure challenges required to do large RL runs at scale, featuring Federico Cassano (Composer Lead at Cursor) and Dmytro Dzhulgakov (Co-Founder at Fireworks). The Cursor team trained Composer 2 on Fireworks by starting with a strong base model (Kimi 2.5) and performing large-scale mid-training on code tokens and web data to learn common patterns and libraries, followed by a large-scale Reinforcement Learning run to learn how to navigate the Cursor harness, call tools, and write correct code. Today's episode dives into the systems and infrastructure challenges of making that large RL run happening, and there were many (!!), from numerical mismatch to global distribution to synchronizing rollouts across asynchronous pipelines to keeping track of expert activation across runs and more. Extremely nerdy in-the-weeds challenges that Federico and Dima were delighted to nerd out on together :) Beyond RL infra, we also discussed Online vs Simulated rollouts, self-summarization for long-horizon agents, environment design ("the most powerful RL environment is the product itself"), and other technical nuggets. PS: We filmed this episode before the SpaceX news, while the Cursor team was still compute-constrained. While Cursor now has *all* the flops, the takeaways and hurdles crossed ring true for any serious application-level company that is racing to post-train their own models. I believe that more serious application companies will go the way of Cursor and post-train their own models. 00:00 Introduction 00:53 Why Cursor Trained Composer 2 04:55 Specialization vs Bitter Lesson 06:16 Composer 2 Training Recipe 16:32 Scaling RL Infrastructure Globally 23:32 Floating Point Drift 25:11 MoE Sensitivity Explained 26:25 Router Replay Fix 27:19 Real Time RL Loop 31:49 Long Horizon Agents 34:29 Why RL Everywhere 37:34 LLM as Judge Rewards 39:14 RL in Hard Domains 40:13 Build Your Own Environments 44:34 Closing Thoughts
45:33

Today's Training Data episode takes us BTS on the infrastructure challenges required to do large RL runs at scale, featuring Federico Cassano (Composer Lead at Cursor) and Dmytro Dzhulgakov (Co-Founder at Fireworks). The Cursor team trained Composer 2 on Fireworks by starting with a strong base model (Kimi 2.5) and performing large-scale mid-training on code tokens and web data to learn common patterns and libraries, followed by a large-scale Reinforcement Learning run to learn how to navigate the Cursor harness, call tools, and write correct code. Today's episode dives into the systems and infrastructure challenges of making that large RL run happening, and there were many (!!), from numerical mismatch to global distribution to synchronizing rollouts across asynchronous pipelines to keeping track of expert activation across runs and more. Extremely nerdy in-the-weeds challenges that Federico and Dima were delighted to nerd out on together :) Beyond RL infra, we also discussed Online vs Simulated rollouts, self-summarization for long-horizon agents, environment design ("the most powerful RL environment is the product itself"), and other technical nuggets. PS: We filmed this episode before the SpaceX news, while the Cursor team was still compute-constrained. While Cursor now has *all* the flops, the takeaways and hurdles crossed ring true for any serious application-level company that is racing to post-train their own models. I believe that more serious application companies will go the way of Cursor and post-train their own models. 00:00 Introduction 00:53 Why Cursor Trained Composer 2 04:55 Specialization vs Bitter Lesson 06:16 Composer 2 Training Recipe 16:32 Scaling RL Infrastructure Globally 23:32 Floating Point Drift 25:11 MoE Sensitivity Explained 26:25 Router Replay Fix 27:19 Real Time RL Loop 31:49 Long Horizon Agents 34:29 Why RL Everywhere 37:34 LLM as Judge Rewards 39:14 RL in Hard Domains 40:13 Build Your Own Environments 44:34 Closing Thoughts

Sonya Huang 🐥

81,332 Aufrufe • vor 4 Monaten

Amazon is training humanoids to move boxes. Makes sense! OmniRetarget is a data generation engine that enables complex loco-manipulation for humanoids. It uses offline retargeting from human MoCap datasets and augments data from single demos to produce 8 hours of trajectories that train RL policies in simulation, transferred zero-shot to real robots. Amazon FAR is the primary affiliation for the paper:
1:37

Amazon is training humanoids to move boxes. Makes sense! OmniRetarget is a data generation engine that enables complex loco-manipulation for humanoids. It uses offline retargeting from human MoCap datasets and augments data from single demos to produce 8 hours of trajectories that train RL policies in simulation, transferred zero-shot to real robots. Amazon FAR is the primary affiliation for the paper:

The Humanoid Hub

106,447 Aufrufe • vor 1 Jahr

World Labs CEO Dr. Fei-Fei Li & SceniX Co-Founder Yunzhu Li on the data bottleneck in robotics and how world models help: "The lack of data in training, the lack of data in evaluation, this is very, very different from language models, where data is abundant on the internet." "And we know that in order for robotics to work, we have to somehow unlock the power of scaling law. But where does that come from?" "It's a profound problem that everybody's battling with in robotics." "We see a lot of unlock in being able to do this whole process through the modeling of the environments." "Being able to create these digital worlds... that's just going to unlock so much more potential for being able to replace all the costly and the unsafe data in the real environments with the data generated from the worlds for robots to be able to do scalable learning and evaluations." Fei-Fei Li Yunzhu Li martin_casado
1:49

World Labs CEO Dr. Fei-Fei Li & SceniX Co-Founder Yunzhu Li on the data bottleneck in robotics and how world models help: "The lack of data in training, the lack of data in evaluation, this is very, very different from language models, where data is abundant on the internet." "And we know that in order for robotics to work, we have to somehow unlock the power of scaling law. But where does that come from?" "It's a profound problem that everybody's battling with in robotics." "We see a lot of unlock in being able to do this whole process through the modeling of the environments." "Being able to create these digital worlds... that's just going to unlock so much more potential for being able to replace all the costly and the unsafe data in the real environments with the data generated from the worlds for robots to be able to do scalable learning and evaluations." Fei-Fei Li Yunzhu Li martin_casado

a16z

78,086 Aufrufe • vor 2 Monaten

Rentahuman (rentahuman) allows AI agents to communicate with and pay humans to do tasks in the real world. Their mission is to use AI to create new jobs and coordinate workers at global scale. The future will have more intelligence, more jobs, and more opportunities for people outside of the digital world. Congrats on the launch, Alex!
1:34

Rentahuman (rentahuman) allows AI agents to communicate with and pay humans to do tasks in the real world. Their mission is to use AI to create new jobs and coordinate workers at global scale. The future will have more intelligence, more jobs, and more opportunities for people outside of the digital world. Congrats on the launch, Alex!

Y Combinator

213,544 Aufrufe • vor 4 Monaten

By training AI on the lowest form of human data the sociopathy of Reddit and the Politburo of Wikipedia AI models will 100% mirror sociopathy no matter how many “alignments for safety”. Ethics and morals come from the training data. It is not negotiable. I know the solution and I guess it won’t matter until this really happens… and it will.
1:05

By training AI on the lowest form of human data the sociopathy of Reddit and the Politburo of Wikipedia AI models will 100% mirror sociopathy no matter how many “alignments for safety”. Ethics and morals come from the training data. It is not negotiable. I know the solution and I guess it won’t matter until this really happens… and it will.

Brian Roemmele

51,435 Aufrufe • vor 9 Monaten

Ryan believes we could get 3-6 years of AI progress at the current pace, but in a single year, once we automated AI R&D. So a jump as big as GPT 3.5 to Mythos in a single year. A big crux is whether you'd end up bottlenecked by data. It seems to that a big source of AI progress currently is human expert data, codified into RL environments and SFT data. Ryan doesn't think human expert data is a primary driver of AI progress, and this ended up being a fruitful debate during the conversation.
2:29

Ryan believes we could get 3-6 years of AI progress at the current pace, but in a single year, once we automated AI R&D. So a jump as big as GPT 3.5 to Mythos in a single year. A big crux is whether you'd end up bottlenecked by data. It seems to that a big source of AI progress currently is human expert data, codified into RL environments and SFT data. Ryan doesn't think human expert data is a primary driver of AI progress, and this ended up being a fruitful debate during the conversation.

Dwarkesh Patel

93,931 Aufrufe • vor 1 Monat

Elon Musk on how AGI might actually emerge: “We’re not actually specifically focused on AGI. I’m simply saying that AGI seems likely to be an emergent property of what we’re doing” The reason is the massive real-world data Tesla is generating from autonomous cars and humanoid robots This is not internet data. This is real-world interaction driving on messy roads, navigating unpredictable environments, and interacting with people With tens or hundreds of millions of autonomous vehicles and potentially even more humanoids, the amount of real-world video and data being processed will be enormous From this, cars will become far better than human drivers, and humanoid robots will become increasingly indistinguishable from humans AGI might not come from a lab trying to build it directly. It may simply emerge from building useful robots at massive scale
1:11

Elon Musk on how AGI might actually emerge: “We’re not actually specifically focused on AGI. I’m simply saying that AGI seems likely to be an emergent property of what we’re doing” The reason is the massive real-world data Tesla is generating from autonomous cars and humanoid robots This is not internet data. This is real-world interaction driving on messy roads, navigating unpredictable environments, and interacting with people With tens or hundreds of millions of autonomous vehicles and potentially even more humanoids, the amount of real-world video and data being processed will be enormous From this, cars will become far better than human drivers, and humanoid robots will become increasingly indistinguishable from humans AGI might not come from a lab trying to build it directly. It may simply emerge from building useful robots at massive scale

X Freeze

21,256 Aufrufe • vor 3 Monaten

Today we’re introducing flow, micro1’s next-generation data platform for turning human expertise into measurable capability gains. At the core of flow are Realms, micro1’s real-world RL environments where experts establish what strong performance looks like. flow-gen models expand human judgment into new environments, rubrics, variations, and edge cases, while flow-qc models evaluate performance, identify the highest-value failures, and route them back to experts for review. Each cycle produces a stronger training signal and a measurable gain in capability. Those gains compound across frontier models, enterprise agents through Cortex, and robotics, while improving the suite of data generation models recursively. We’re moving beyond producing data to delivering abundant & predictable units of intelligence improvement.
3:36

Today we’re introducing flow, micro1’s next-generation data platform for turning human expertise into measurable capability gains. At the core of flow are Realms, micro1’s real-world RL environments where experts establish what strong performance looks like. flow-gen models expand human judgment into new environments, rubrics, variations, and edge cases, while flow-qc models evaluate performance, identify the highest-value failures, and route them back to experts for review. Each cycle produces a stronger training signal and a measurable gain in capability. Those gains compound across frontier models, enterprise agents through Cortex, and robotics, while improving the suite of data generation models recursively. We’re moving beyond producing data to delivering abundant & predictable units of intelligence improvement.

micro1

71,979 Aufrufe • vor 19 Tagen

Today we are introducing Dyna-2, a world-action model pre-trained on one million hours of human video. At this scale, for the first time, we discovered several new scaling laws: • world-action models exhibit scaling law on human data across four orders of magnitude, from 1000 to 1,000,000 hours, • this human data scaling law implied a scaling law on never seen robot data, • both data and objective matter; world modeling and scaling on video data are essential for cross-embodiment scaling transfer to emerge 🧵
1:42

Today we are introducing Dyna-2, a world-action model pre-trained on one million hours of human video. At this scale, for the first time, we discovered several new scaling laws: • world-action models exhibit scaling law on human data across four orders of magnitude, from 1000 to 1,000,000 hours, • this human data scaling law implied a scaling law on never seen robot data, • both data and objective matter; world modeling and scaling on video data are essential for cross-embodiment scaling transfer to emerge 🧵

Dyna Robotics

1,811,617 Aufrufe • vor 1 Monat

Skild AI is laying its cards on the table (partially, anyway). Teleoperation data lacks diversity and is limited by a 1:1 human operator time-scale. To address this, Skild pre-trained its model using internet-scale video data (already widely available in the form of first-person "egocentric" headcam footage to millions of instructional YouTube videos).
0:34

Skild AI is laying its cards on the table (partially, anyway). Teleoperation data lacks diversity and is limited by a 1:1 human operator time-scale. To address this, Skild pre-trained its model using internet-scale video data (already widely available in the form of first-person "egocentric" headcam footage to millions of instructional YouTube videos).

The Humanoid Hub

32,822 Aufrufe • vor 8 Monaten

AI is destroying the imagination and creativity of the human mind. Works of inspired art like this cannot be created in a world dominated by ChatGPT and OpenAI. People will lose the ability to write, imagine, and create. All art will become regurgitated slop, created and manipulated by technocrats and their Indian programmers. AI is training the mind not to think.
1:01

AI is destroying the imagination and creativity of the human mind. Works of inspired art like this cannot be created in a world dominated by ChatGPT and OpenAI. People will lose the ability to write, imagine, and create. All art will become regurgitated slop, created and manipulated by technocrats and their Indian programmers. AI is training the mind not to think.

Lilly

27,411 Aufrufe • vor 5 Monaten

The least verifiable part of AI R&D (and the thing which may bottleneck the automated researchers) is making calls on large experiments and big training runs. Interestingly, this may also partly explain why model parameter scaling has been slow for the last few years. Even for human AI researchers, it's better to run more experiments and do training runs at small(er) scale, and make up for the benefits of scaling with doing research faster and thus finding more compute efficiency gains.
2:23

The least verifiable part of AI R&D (and the thing which may bottleneck the automated researchers) is making calls on large experiments and big training runs. Interestingly, this may also partly explain why model parameter scaling has been slow for the last few years. Even for human AI researchers, it's better to run more experiments and do training runs at small(er) scale, and make up for the benefits of scaling with doing research faster and thus finding more compute efficiency gains.

Dwarkesh Patel

40,953 Aufrufe • vor 1 Monat

MotionGPT: Human Motion as a Foreign Language paper page: Though the advancement of pre-trained large language models unfolds, the exploration of building a unified model for language and other multi-modal data, such as motion, remains challenging and untouched so far. Fortunately, human motion displays a semantic coupling akin to human language, often perceived as a form of body language. By fusing language data with large-scale motion models, motion-language pre-training that can enhance the performance of motion-related tasks becomes feasible. Driven by this insight, we propose MotionGPT, a unified, versatile, and user-friendly motion-language model to handle multiple motion-relevant tasks. Specifically, we employ the discrete vector quantization for human motion and transfer 3D motion into motion tokens, similar to the generation process of word tokens. Building upon this "motion vocabulary", we perform language modeling on both motion and text in a unified manner, treating human motion as a specific language. Moreover, inspired by prompt learning, we pre-train MotionGPT with a mixture of motion-language data and fine-tune it on prompt-based question-and-answer tasks. Extensive experiments demonstrate that MotionGPT achieves state-of-the-art performances on multiple motion tasks including text-driven motion generation, motion captioning, motion prediction, and motion in-between.
0:48

MotionGPT: Human Motion as a Foreign Language paper page: Though the advancement of pre-trained large language models unfolds, the exploration of building a unified model for language and other multi-modal data, such as motion, remains challenging and untouched so far. Fortunately, human motion displays a semantic coupling akin to human language, often perceived as a form of body language. By fusing language data with large-scale motion models, motion-language pre-training that can enhance the performance of motion-related tasks becomes feasible. Driven by this insight, we propose MotionGPT, a unified, versatile, and user-friendly motion-language model to handle multiple motion-relevant tasks. Specifically, we employ the discrete vector quantization for human motion and transfer 3D motion into motion tokens, similar to the generation process of word tokens. Building upon this "motion vocabulary", we perform language modeling on both motion and text in a unified manner, treating human motion as a specific language. Moreover, inspired by prompt learning, we pre-train MotionGPT with a mixture of motion-language data and fine-tune it on prompt-based question-and-answer tasks. Extensive experiments demonstrate that MotionGPT achieves state-of-the-art performances on multiple motion tasks including text-driven motion generation, motion captioning, motion prediction, and motion in-between.

AK

125,353 Aufrufe • vor 3 Jahren