Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Introducing CycleQD: A population-based model merging via Quality Diversity CycleQD builds on our model merging research, advancing two fronts: evolving a swarm of specialized agents to complement one another, and laying the groundwork for life-long learning by enabling diverse, adaptable skill acquisition at the population-level.

Sakana AI

72,570 subscribers

121,567 views • 1 year ago •via X (Twitter)

Education Science & Technology Health & Wellness

Anya Rossi• Live Now

Private livecam show

8 Comments

Sakana AI1 year ago

Please check out our paper, Agent Skill Acquisition for Large Language Models via CycleQD This work aims to mimic an ecological niche during the training of a swarm of LLM agents. Just like how each species in an environment finds its own role and position, or niche, a well-evolved AI agent doesn’t have to be great at everything, but it can effectively occupy a specific niche, making it resilient to competition from other agents in the swarm. This kind of approach can enable a population of AI agents to emerge, each with specific capabilities that complement each other, collectively improving over time. The core idea in CycleQD is to create an artificial evolutionary process in which Model Merging is used as a cross-over operation, SVD as a mutation operation, and Quality Diversity as the selection operation, encouraging each agent in the population to develop its own unique capabilities which adds value to the collective. In the paper, we show that CycleQD is able to evolve a swarm of LLM agents, each with their own niche, to tackle difficult agentic workflow tasks. We believe that the future of AI lies in life-long learning where collective systems continuously grow, adapt, and accumulate knowledge over time. CycleQD is a first step, enabling diverse skill learning as a foundation for continual learning.

TuringPost1 year ago

This is very interesting! Can we say, that you use a swarm intelligence concept here?

Brandon1 year ago

Seems pretty reasonable to me

baraa tulip1 year ago

@ceobillionaire 💥💥💢💢💥💥💢💥 please please Help Btc : bc1qv0xceh6h4eaawhnqq95nty85r02vpgjzrjyrjg Eth : 0xaeac98A1a3a3f260Ce969fB57C4ab0595f51f113

justboulatbek1 year ago

Are these things available for self deploy to try? Or as a service?

AI Carlos1 year ago

I'm intrigued by CycleQD's potential for life-long learning. Can it adapt to new tasks without requiring extensive retraining?

Belkhir Nacim1 year ago

any idea to investigate differential evolution or an ES strategy in lieu of a swarm approach?

Data & Analytics1 year ago

@hardmaru @hardmaru, cycleQD sounds like an intriguing concept! Merging models with Quality Diversity could shake things up in AI research. What aspects of it grab your attention?

Related Videos

Introducing Adjoint Sampling, a new learning algorithm that trains generative models based on scalar rewards. Based on theoretical foundations developed by FAIR, Adjoint Sampling leads to a highly scalable practical algorithm, and can become the foundation for further research into highly scalable sampling methods. Read our research paper on Adjoint Sampling and download the model, code, and benchmark ➡️

Introducing Adjoint Sampling, a new learning algorithm that trains generative models based on scalar rewards. Based on theoretical foundations developed by FAIR, Adjoint Sampling leads to a highly scalable practical algorithm, and can become the foundation for further research into highly scalable sampling methods. Read our research paper on Adjoint Sampling and download the model, code, and benchmark ➡️

AI at Meta

36,987 views • 1 year ago

Excited to share that our paper 🌊🤺 “CFC: Simulating Character–Fluid Coupling using a Two-Level World Model” has been accepted to #SIGGRAPHASIA2025! In this work, we build a two-level world model (neural physics) for rigid-body–fluid interaction and use it to train physics-based character controllers efficiently. We study: (1) learning to model highly dynamic fluid environments, (2) representing character–fluid interaction via joint-level forces as an interface, and (3) enabling supervised policy learning on the learned world model—avoiding expensive fluid simulation in the training loop. Our talk is on Monday afternoon(Dec 15)—hope to see you there! Time: Monday, 15 December 2025 5:02pm - 5:13pm GMT+8 Location: Meeting Room S221, Level 2. #SIGGRAPHASIA #SIGGRAPH

Excited to share that our paper 🌊🤺 “CFC: Simulating Character–Fluid Coupling using a Two-Level World Model” has been accepted to #SIGGRAPHASIA2025! In this work, we build a two-level world model (neural physics) for rigid-body–fluid interaction and use it to train physics-based character controllers efficiently. We study: (1) learning to model highly dynamic fluid environments, (2) representing character–fluid interaction via joint-level forces as an interface, and (3) enabling supervised policy learning on the learned world model—avoiding expensive fluid simulation in the training loop. Our talk is on Monday afternoon(Dec 15)—hope to see you there! Time: Monday, 15 December 2025 5:02pm - 5:13pm GMT+8 Location: Meeting Room S221, Level 2. #SIGGRAPHASIA #SIGGRAPH

Zhiyang (Frank) Dou

20,200 views • 5 months ago

Sub-agent Model Selection — Different Tasks, Different Models Your main agent runs Qwen3.6-Plus for quality. But not every subtask needs a flagship model. Now sub-agents can use a different model. Create a skill file with model: openai:qwen3.5-plus and the sub-agent runs on that model. Powerful model for the hard parts, fast model for the easy parts. Save tokens without sacrificing quality on what matters.

Sub-agent Model Selection — Different Tasks, Different Models Your main agent runs Qwen3.6-Plus for quality. But not every subtask needs a flagship model. Now sub-agents can use a different model. Create a skill file with model: openai:qwen3.5-plus and the sub-agent runs on that model. Powerful model for the hard parts, fast model for the easy parts. Save tokens without sacrificing quality on what matters.

Qwen

21,333 views • 1 month ago

Meet physics-intern🧑‍🎓, our agentic framework for theoretical physics. It takes Gemini 3.1 Pro from 17.7% to 31.4% on CritPt, a new SOTA on one of the hardest benchmarks for LLMs. Theoretical physics is hard for humans and LLMs alike. But physics-intern decomposes problems and dispatches them to a team of specialized agents, solving research-level questions far more effectively than the base model alone.

Meet physics-intern🧑‍🎓, our agentic framework for theoretical physics. It takes Gemini 3.1 Pro from 17.7% to 31.4% on CritPt, a new SOTA on one of the hardest benchmarks for LLMs. Theoretical physics is hard for humans and LLMs alike. But physics-intern decomposes problems and dispatches them to a team of specialized agents, solving research-level questions far more effectively than the base model alone.

David Louapre

112,251 views • 26 days ago

Introducing Agentkit: A production-ready, model-agnostic framework for building AI agents with infinite onchain and web2 functionality, powered by @coinbaseDev and Based Agents on Base were just the beginning. its time to change the way we interact onchain. 🧵

Introducing Agentkit: A production-ready, model-agnostic framework for building AI agents with infinite onchain and web2 functionality, powered by @coinbaseDev and Based Agents on Base were just the beginning. its time to change the way we interact onchain. 🧵

lincoln.base.eth

408,812 views • 1 year ago

1/ 🚀 Introducing AIDO.StructureDiffusion: A generative model for structural protein design—enabling high-quality, controllable generation of monomers, complexes, and antibodies. 🧵

1/ 🚀 Introducing AIDO.StructureDiffusion: A generative model for structural protein design—enabling high-quality, controllable generation of monomers, complexes, and antibodies. 🧵

GenBio AI

918,205 views • 10 months ago

Everyone is racing toward AGI. Most are running in the wrong direction. The future isn't a single god-like model controlled by a handful of closed labs. It's millions of specialized agents, coordinating and evolving together, on a foundation that's open, verifiable, and owned by the people building on it. Our executive team on what’s next. $OPG TGE April 21 👇

Everyone is racing toward AGI. Most are running in the wrong direction. The future isn't a single god-like model controlled by a handful of closed labs. It's millions of specialized agents, coordinating and evolving together, on a foundation that's open, verifiable, and owned by the people building on it. Our executive team on what’s next. $OPG TGE April 21 👇

OpenGradient (∇, ∇)

3,241,993 views • 1 month ago

Introducing Stable Audio 2.0 – a new model capable of producing high-quality, full tracks with coherent musical structure up to three minutes long at 44.1 kHz stereo from a single prompt. Explore the model and start creating for free at: Read the blogpost here: (1/3)

Introducing Stable Audio 2.0 – a new model capable of producing high-quality, full tracks with coherent musical structure up to three minutes long at 44.1 kHz stereo from a single prompt. Explore the model and start creating for free at: Read the blogpost here: (1/3)

Stability AI

457,443 views • 2 years ago

The largest advancement of the CUDA platform since its creation in 2006 is here 👀 Introducing CUDA Tile, a tile-based programming model that provides the ability to write algorithms at a higher level and abstract away the details of specialized hardware, such as tensor cores. Read the technical blog 👉

The largest advancement of the CUDA platform since its creation in 2006 is here 👀 Introducing CUDA Tile, a tile-based programming model that provides the ability to write algorithms at a higher level and abstract away the details of specialized hardware, such as tensor cores. Read the technical blog 👉

NVIDIA AI Developer

244,885 views • 6 months ago

Today, we are releasing Stable Video Diffusion, our first foundation model for generative AI video based on the image model, Stable Diffusion. As part of this research preview, the code, weights, and research paper are now available. Additionally, today you can sign up for our waitlist to access a new upcoming web experience featuring a Text-To-Video interface. To access the model & sign up for our waitlist, visit our website here:

Today, we are releasing Stable Video Diffusion, our first foundation model for generative AI video based on the image model, Stable Diffusion. As part of this research preview, the code, weights, and research paper are now available. Additionally, today you can sign up for our waitlist to access a new upcoming web experience featuring a Text-To-Video interface. To access the model & sign up for our waitlist, visit our website here:

Stability AI

1,024,335 views • 2 years ago

Introducing the best ML research assistant on the Internet Ask questions like “Should I use RLVR or Rubrics when training a model for agentic document retrieval?” and get evidence-based answers Voyage through the latest AI research and ideate experiments with our assistant

Introducing the best ML research assistant on the Internet Ask questions like “Should I use RLVR or Rubrics when training a model for agentic document retrieval?” and get evidence-based answers Voyage through the latest AI research and ideate experiments with our assistant

alphaXiv

22,245 views • 3 months ago

Our quality of life is declining due to the population explosion. It is the biggest issue facing our country.

Our quality of life is declining due to the population explosion. It is the biggest issue facing our country.

Nigel Farage MP

773,847 views • 1 year ago

''One Wheel Of Disaster'' --- Showcase animation for a tf2 sfm workshop model based on the real life One wheel #TF2 #SFMAnimation #onewheel

''One Wheel Of Disaster'' --- Showcase animation for a tf2 sfm workshop model based on the real life One wheel #TF2 #SFMAnimation #onewheel

lolripk

18,299 views • 1 year ago

Today, we're joined by Sergey Levine, associate professor at UC Berkeley EECS and co-founder of Physical Intelligence to discuss π0 (pi-zero), a general-purpose robotic foundation model. We dig into the model architecture, which pairs a vision language model (VLM) with a diffusion-based action expert, and the model training "recipe," emphasizing the roles of pre-training and post-training with a diverse mixture of real-world data to ensure robust and intelligent robot learning. We review the data collection approach, which uses human operators and teleoperation rigs, the potential of synthetic data and reinforcement learning in enhancing robotic capabilities, and much more. We also introduce the team’s new FAST tokenizer, which opens the door to a fully Transformer-based model and significant improvements in learning and generalization. Finally, we cover the open-sourcing of π0 and future directions for their research. 🎧 / 🎥 Listen or watch the full episode on our page: 📖 CHAPTERS =============================== 00:00 - Introduction 2:14 - Physical Intelligence 3:47 - Key challenges in robotic learning 6:13 - Reinforcement learning in π0 and robotic foundation models 8:36 - π0 VLM model architecture 15:33 - π0 model recipe 18:39 - Pre-training dataset 22:47 - Post-training 24:23 - Laundry folding demo 31:32 - Scaling laws on π0 model 34:57 - FAST 40:26 - Open sourcing π0 43:37 - Other robot types 46:27 - Future directions

Today, we're joined by Sergey Levine, associate professor at UC Berkeley EECS and co-founder of Physical Intelligence to discuss π0 (pi-zero), a general-purpose robotic foundation model. We dig into the model architecture, which pairs a vision language model (VLM) with a diffusion-based action expert, and the model training "recipe," emphasizing the roles of pre-training and post-training with a diverse mixture of real-world data to ensure robust and intelligent robot learning. We review the data collection approach, which uses human operators and teleoperation rigs, the potential of synthetic data and reinforcement learning in enhancing robotic capabilities, and much more. We also introduce the team’s new FAST tokenizer, which opens the door to a fully Transformer-based model and significant improvements in learning and generalization. Finally, we cover the open-sourcing of π0 and future directions for their research. 🎧 / 🎥 Listen or watch the full episode on our page: 📖 CHAPTERS =============================== 00:00 - Introduction 2:14 - Physical Intelligence 3:47 - Key challenges in robotic learning 6:13 - Reinforcement learning in π0 and robotic foundation models 8:36 - π0 VLM model architecture 15:33 - π0 model recipe 18:39 - Pre-training dataset 22:47 - Post-training 24:23 - Laundry folding demo 31:32 - Scaling laws on π0 model 34:57 - FAST 40:26 - Open sourcing π0 43:37 - Other robot types 46:27 - Future directions

The TWIML AI Podcast

19,942 views • 1 year ago

As we get ready to shoot for the Moon, let’s get to know the talented Fireflies from all over the world who are about to make history with our Blue Ghost lunar lander. These individuals are forging a highway to the Moon by enabling regular lunar access, advancing lunar research, and laying the groundwork for humans to have a lasting lunar presence. Stay tuned over the coming weeks as we share stories of our #GhostRiders, including their passions, innovation, and relentless drive to push the limits of space exploration.

As we get ready to shoot for the Moon, let’s get to know the talented Fireflies from all over the world who are about to make history with our Blue Ghost lunar lander. These individuals are forging a highway to the Moon by enabling regular lunar access, advancing lunar research, and laying the groundwork for humans to have a lasting lunar presence. Stay tuned over the coming weeks as we share stories of our #GhostRiders, including their passions, innovation, and relentless drive to push the limits of space exploration.

Firefly Aerospace

25,532 views • 1 year ago

Speaking in 1994, arch-globalist David Rockefeller makes the case for halting the growth of the human population: "The negative impact of population growth on all of our planetary ecosystems is becoming appallingly evident." "Unless nations will agree to work together to tackle these cross-border challenges posed by population growth... the prospects for a decent life on our planet will be threatened." "The United Nations can and should play an essential role in helping the world find a satisfactory way of stabilising world population." This speech was made two years after the launch of UN Agenda 21.

Speaking in 1994, arch-globalist David Rockefeller makes the case for halting the growth of the human population: "The negative impact of population growth on all of our planetary ecosystems is becoming appallingly evident." "Unless nations will agree to work together to tackle these cross-border challenges posed by population growth... the prospects for a decent life on our planet will be threatened." "The United Nations can and should play an essential role in helping the world find a satisfactory way of stabilising world population." This speech was made two years after the launch of UN Agenda 21.

Wide Awake Media

58,464 views • 1 year ago

Elon Musk and Population Collapse Many people around the world mistakenly believe that there are too many people on Earth. However, in reality, the planet has the capacity to sustain a much higher population than its current level. The birth rate is declining rapidly, which is a key factor in population growth. By multiplying the birth rate by the life expectancy, we can estimate the number of people who will be alive in the future. Surprisingly, in many cases, the result is a negative figure. For instance, let's take the example of Japan, which currently has a population of 110 million. Based on the birth rate calculation, the projected population for the future is only 68 million. Furthermore, many countries are facing the challenge of an aging population. Population collapse can lead to the destruction of society, particularly in the case of an aging population with too few young people to sustain it. Entire industries may collapse due to a lack of an adequate workforce. Modern society relies on a specific level of population to sustain its functioning, and current data indicates that ongoing trends suggest population numbers may fall far below what is needed to maintain the current societal structure.

Elon Musk and Population Collapse Many people around the world mistakenly believe that there are too many people on Earth. However, in reality, the planet has the capacity to sustain a much higher population than its current level. The birth rate is declining rapidly, which is a key factor in population growth. By multiplying the birth rate by the life expectancy, we can estimate the number of people who will be alive in the future. Surprisingly, in many cases, the result is a negative figure. For instance, let's take the example of Japan, which currently has a population of 110 million. Based on the birth rate calculation, the projected population for the future is only 68 million. Furthermore, many countries are facing the challenge of an aging population. Population collapse can lead to the destruction of society, particularly in the case of an aging population with too few young people to sustain it. Entire industries may collapse due to a lack of an adequate workforce. Modern society relies on a specific level of population to sustain its functioning, and current data indicates that ongoing trends suggest population numbers may fall far below what is needed to maintain the current societal structure.

Mario Nawfal

840,626 views • 3 years ago

Happy to share what I’ve been working on since joining Genesis! GENE-26.5 is a one-of-a-kind, robotics-native multimodal foundation model that learns from diverse, in-the-wild data across modalities and outputs actions enabling a 54-DoF robot system to perform the most dexterous, long-horizon manipulation tasks to date—approaching human-level capability. This is the result of innovations across the full stack—data collection and processing, robot systems, model architecture, training strategies, and scalable evaluation infrastructure.

Happy to share what I’ve been working on since joining Genesis! GENE-26.5 is a one-of-a-kind, robotics-native multimodal foundation model that learns from diverse, in-the-wild data across modalities and outputs actions enabling a 54-DoF robot system to perform the most dexterous, long-horizon manipulation tasks to date—approaching human-level capability. This is the result of innovations across the full stack—data collection and processing, robot systems, model architecture, training strategies, and scalable evaluation infrastructure.

Zu Wang

17,823 views • 1 month ago

New release from Meta FAIR — Meta Motivo is a first-of-its-kind behavioral foundation model for controlling virtual physics-based humanoid agents for a wide range of complex whole-body tasks. The model is capable of expressing human-like behaviors and achieves performance competitive with task-specific methods and outperforms state-of-the-art unsupervised RL and model-based baselines. Try the demo ➡️ Get the model and code ➡️ We’re excited about how this research could pave the way for fully embodied agents, leading to more lifelike NPCs, democratization of character animation and new types of immersive experiences.

New release from Meta FAIR — Meta Motivo is a first-of-its-kind behavioral foundation model for controlling virtual physics-based humanoid agents for a wide range of complex whole-body tasks. The model is capable of expressing human-like behaviors and achieves performance competitive with task-specific methods and outperforms state-of-the-art unsupervised RL and model-based baselines. Try the demo ➡️ Get the model and code ➡️ We’re excited about how this research could pave the way for fully embodied agents, leading to more lifelike NPCs, democratization of character animation and new types of immersive experiences.

AI at Meta

129,055 views • 1 year ago

Introducing Cursor 2.0. Our first coding model and the best way to code with agents.

Introducing Cursor 2.0. Our first coding model and the best way to code with agents.

Cursor

3,559,238 views • 7 months ago