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.

Announcing EXO Gym: Simulate distributed training environments using just your laptop. Previously, distributed training experiments required setting up complex multi-node clusters. With EXO Gym, multiple virtual nodes are spawned within one device. 🧵

Matt Beton

2,498 subscribers

51,365 views • 11 months ago •via X (Twitter)

Science & Technology Education

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

OptimAI Core Node v2 is live. And with it, the introduction of OptimAI Claw. From nodes as infrastructure → to nodes as execution environments. With Core Node v2, every device becomes a runtime for native personal agents. 👉 Get your node up and running. No complex setup. No fragmented tooling. Just a few clicks... and your agent is live. OptimAI Claw brings persistent, node-resident agents directly into your system: • Multi-source research across web & social • Real-time summarization and signal tracking • Autonomous workflow execution • Continuous learning through reinforcement All running locally. All under your control. Supported on Windows & macOS. This is beyond prompt-based AI. Beyond isolated tools. This is personal agent infrastructure. Where every node thinks, acts, and evolves. Claw is now part of the network. And this is only the beginning. Happy Building, Making Impact! 🔥

OptimAI Core Node v2 is live. And with it, the introduction of OptimAI Claw. From nodes as infrastructure → to nodes as execution environments. With Core Node v2, every device becomes a runtime for native personal agents. 👉 Get your node up and running. No complex setup. No fragmented tooling. Just a few clicks... and your agent is live. OptimAI Claw brings persistent, node-resident agents directly into your system: • Multi-source research across web & social • Real-time summarization and signal tracking • Autonomous workflow execution • Continuous learning through reinforcement All running locally. All under your control. Supported on Windows & macOS. This is beyond prompt-based AI. Beyond isolated tools. This is personal agent infrastructure. Where every node thinks, acts, and evolves. Claw is now part of the network. And this is only the beginning. Happy Building, Making Impact! 🔥

OptimAI Network

333,727 views • 4 months ago

Author of Designing Data-Intensive Applications, Martin Kleppmann, gives us a peek into "The Troubles with Distributed Systems" and why you can't assume things behave well in distributed systems: "The whole idea is that in distributed system theory, there are certain things that we tend to assume. For example, we just assume that there's no upper bound on how long it might take for a message to go over the network. So when you send a message, it might arrive within a hundred microseconds, or it might take 10 years, and distributed system theory just doesn't make any assumptions about that sort of timing if we can avoid it. Or rather, some theory does make those assumptions, but it's a dangerous assumption to make because occasionally the network delay does become much higher than what is typical. Another thing is about crashes. Distributed system theory just says nodes can crash, but what does that actually mean? What in practice does it mean for a node to become unavailable? Because it might be a software crash, but it might be a hardware failure. It might be somebody unplugging the power cable. It might be that the node is actually still running, but it's just become disconnected from the network. And so, the point of this book chapter really is to defend and justify the theoretical models that we use for analysing distributed systems and to give a lot of stories and case studies to show that, actually, tonnes of stuff does go wrong. Don't believe anyone who says, 'Oh, failures are rare. Don't worry about it. It's fine’. Actually, no. If you want to make things reliable, you really do have to worry about a whole bunch of weird, unusual, but certainly possible edge cases. Timing is another one of those things. It's very easy to assume that your clocks are correct, and most of the time the clocks are pretty correct, but we just can't rely on it because actually they're just not precise enough on the whole. It's very tempting to make certain assumptions that things are well behaved and in distributed systems, we just have to try to get away from those assumptions if we want the systems to work reliably, even in the face of things going wrong."

Author of Designing Data-Intensive Applications, Martin Kleppmann, gives us a peek into "The Troubles with Distributed Systems" and why you can't assume things behave well in distributed systems: "The whole idea is that in distributed system theory, there are certain things that we tend to assume. For example, we just assume that there's no upper bound on how long it might take for a message to go over the network. So when you send a message, it might arrive within a hundred microseconds, or it might take 10 years, and distributed system theory just doesn't make any assumptions about that sort of timing if we can avoid it. Or rather, some theory does make those assumptions, but it's a dangerous assumption to make because occasionally the network delay does become much higher than what is typical. Another thing is about crashes. Distributed system theory just says nodes can crash, but what does that actually mean? What in practice does it mean for a node to become unavailable? Because it might be a software crash, but it might be a hardware failure. It might be somebody unplugging the power cable. It might be that the node is actually still running, but it's just become disconnected from the network. And so, the point of this book chapter really is to defend and justify the theoretical models that we use for analysing distributed systems and to give a lot of stories and case studies to show that, actually, tonnes of stuff does go wrong. Don't believe anyone who says, 'Oh, failures are rare. Don't worry about it. It's fine’. Actually, no. If you want to make things reliable, you really do have to worry about a whole bunch of weird, unusual, but certainly possible edge cases. Timing is another one of those things. It's very easy to assume that your clocks are correct, and most of the time the clocks are pretty correct, but we just can't rely on it because actually they're just not precise enough on the whole. It's very tempting to make certain assumptions that things are well behaved and in distributed systems, we just have to try to get away from those assumptions if we want the systems to work reliably, even in the face of things going wrong."

The Pragmatic Engineer

50,567 views • 3 months ago

A new robot policy just cleaned up a kitchen it had never seen before [watch what happens. paper included ⬇️] Pi-0.5 builds on top of Pi-0 and shows how smart co-training with diverse data can unlock real generalization in the home. It doesn’t just learn from one setup but adapts to many, including homes it’s never seen. What it does ✅ Handles new homes without training in them ✅ Follows complex language instructions ✅ Cleans, places dishes, handles spills ✅ Matches in-home training models using cross-embodiment and web data Robots that understand tasks and adapt to new spaces are finally within reach. More in the blog: Read the paper ⬇️ Physical Intelligence, co-founded by UC Berkeley professor Sergey Levine, is a robotics startup developing general-purpose AI foundation models that enable robots to perform a wide variety of real-world tasks with human-like adaptability, recently raising $400 million to advance this vision.

A new robot policy just cleaned up a kitchen it had never seen before [watch what happens. paper included ⬇️] Pi-0.5 builds on top of Pi-0 and shows how smart co-training with diverse data can unlock real generalization in the home. It doesn’t just learn from one setup but adapts to many, including homes it’s never seen. What it does ✅ Handles new homes without training in them ✅ Follows complex language instructions ✅ Cleans, places dishes, handles spills ✅ Matches in-home training models using cross-embodiment and web data Robots that understand tasks and adapt to new spaces are finally within reach. More in the blog: Read the paper ⬇️ Physical Intelligence, co-founded by UC Berkeley professor Sergey Levine, is a robotics startup developing general-purpose AI foundation models that enable robots to perform a wide variety of real-world tasks with human-like adaptability, recently raising $400 million to advance this vision.

Ilir Aliu - eu/acc

18,120 views • 1 year ago

𝐈'𝐯𝐞 𝐟𝐨𝐮𝐧𝐝 Vultisig 𝐭𝐨 𝐛𝐞 𝐭𝐡𝐞 𝐛𝐞𝐬𝐭 𝐜𝐡𝐨𝐢𝐜𝐞. 𝐇𝐞𝐫𝐞'𝐬 𝐦𝐲 𝐭𝐨𝐩 𝐚𝐝𝐯𝐢𝐜𝐞 𝐟𝐨𝐫 𝐦𝐚𝐱𝐢𝐦𝐢𝐳𝐢𝐧𝐠 𝐬𝐞𝐜𝐮𝐫𝐢𝐭𝐲 𝐚𝐧𝐝 𝐞𝐚𝐬𝐞 𝐨𝐟 𝐮𝐬𝐞: 𝐖𝐡𝐚𝐭’𝐬 𝐕𝐮𝐥𝐭𝐢𝐬𝐢𝐠? Vultisig is a multi-chain multi-signature wallet, which means it's like your bank's 2FA for blockchain transactions. Vultisig enables you to customize your security level based on your needs. You can set up transactions on your device and use another device to approve them. For an everyday wallet, set up a Fast Vault on a single device like your phone using Vultiserver. This allows you to make transactions swiftly without needing multiple devices to sign in. If you want a more secure yet accessible wallet, create a 2-of-3 Secure Vault using devices you have at least two of (e.g., laptop, tablet, and phone). Protect long-term assets by setting up a 3-of-4 Secure Vault with separate devices, minimizing the chances of theft or unauthorized access to the minimum required for signing. 𝐌𝐚𝐱𝐢𝐦𝐢𝐳𝐞 𝐕𝐮𝐥𝐭𝐢𝐬𝐢𝐠 𝐖𝐚𝐥𝐥𝐞𝐭 𝐒𝐞𝐜𝐮𝐫𝐢𝐭𝐲 𝐰𝐢𝐭𝐡 𝐓𝐡𝐞𝐬𝐞 𝐓𝐨𝐩 𝐓𝐢𝐩𝐬 With numerous options and settings available, it's essential to follow best practices for optimal security. Keep these three key tips in mind: 𝐏𝐫𝐨𝐭𝐞𝐜𝐭 𝐘𝐨𝐮𝐫 𝐕𝐮𝐥𝐭𝐢𝐬𝐢𝐠 𝐖𝐚𝐥𝐥𝐞𝐭 𝐁𝐚𝐜𝐤𝐮𝐩𝐬: ➛ Secure multiple backups from each device and store them separately to avoid single points of failure. Keep backups in secure locations, such as encrypted cloud storage or separate, off-site devices like memory sticks or portable hard drives. ➛ Once your wallet is set up and all backups are made, remove the Vault Shares from every device and restore them from your backups. This ensures your backups are functioning and error-free. ➛ Before transferring significant funds, ensure your new wallet functions correctly by sending a small amount back to your old wallet. Once all checks are complete, you're ready to embrace the Vultisig experience! Do you have vulties? Click here to get started ☟ Join Vultisig dc to always get latest updates on Vultisig☟ Remember that anyone with enough backups to meet the signing threshold can access your funds. Safeguard backup files with strong passwords and store them across different devices and locations for maximum security. Please watch the video below to understand Vultisig better. If you have questions don't hesitate to leave a comment below, and I'll be happy to reply! 😊

𝐈'𝐯𝐞 𝐟𝐨𝐮𝐧𝐝 Vultisig 𝐭𝐨 𝐛𝐞 𝐭𝐡𝐞 𝐛𝐞𝐬𝐭 𝐜𝐡𝐨𝐢𝐜𝐞. 𝐇𝐞𝐫𝐞'𝐬 𝐦𝐲 𝐭𝐨𝐩 𝐚𝐝𝐯𝐢𝐜𝐞 𝐟𝐨𝐫 𝐦𝐚𝐱𝐢𝐦𝐢𝐳𝐢𝐧𝐠 𝐬𝐞𝐜𝐮𝐫𝐢𝐭𝐲 𝐚𝐧𝐝 𝐞𝐚𝐬𝐞 𝐨𝐟 𝐮𝐬𝐞: 𝐖𝐡𝐚𝐭’𝐬 𝐕𝐮𝐥𝐭𝐢𝐬𝐢𝐠? Vultisig is a multi-chain multi-signature wallet, which means it's like your bank's 2FA for blockchain transactions. Vultisig enables you to customize your security level based on your needs. You can set up transactions on your device and use another device to approve them. For an everyday wallet, set up a Fast Vault on a single device like your phone using Vultiserver. This allows you to make transactions swiftly without needing multiple devices to sign in. If you want a more secure yet accessible wallet, create a 2-of-3 Secure Vault using devices you have at least two of (e.g., laptop, tablet, and phone). Protect long-term assets by setting up a 3-of-4 Secure Vault with separate devices, minimizing the chances of theft or unauthorized access to the minimum required for signing. 𝐌𝐚𝐱𝐢𝐦𝐢𝐳𝐞 𝐕𝐮𝐥𝐭𝐢𝐬𝐢𝐠 𝐖𝐚𝐥𝐥𝐞𝐭 𝐒𝐞𝐜𝐮𝐫𝐢𝐭𝐲 𝐰𝐢𝐭𝐡 𝐓𝐡𝐞𝐬𝐞 𝐓𝐨𝐩 𝐓𝐢𝐩𝐬 With numerous options and settings available, it's essential to follow best practices for optimal security. Keep these three key tips in mind: 𝐏𝐫𝐨𝐭𝐞𝐜𝐭 𝐘𝐨𝐮𝐫 𝐕𝐮𝐥𝐭𝐢𝐬𝐢𝐠 𝐖𝐚𝐥𝐥𝐞𝐭 𝐁𝐚𝐜𝐤𝐮𝐩𝐬: ➛ Secure multiple backups from each device and store them separately to avoid single points of failure. Keep backups in secure locations, such as encrypted cloud storage or separate, off-site devices like memory sticks or portable hard drives. ➛ Once your wallet is set up and all backups are made, remove the Vault Shares from every device and restore them from your backups. This ensures your backups are functioning and error-free. ➛ Before transferring significant funds, ensure your new wallet functions correctly by sending a small amount back to your old wallet. Once all checks are complete, you're ready to embrace the Vultisig experience! Do you have vulties? Click here to get started ☟ Join Vultisig dc to always get latest updates on Vultisig☟ Remember that anyone with enough backups to meet the signing threshold can access your funds. Safeguard backup files with strong passwords and store them across different devices and locations for maximum security. Please watch the video below to understand Vultisig better. If you have questions don't hesitate to leave a comment below, and I'll be happy to reply! 😊

𝗡𝗜𝗖𝗢𝗟𝗘✰

26,645 views • 1 year ago

Cloud GPU training is a scam. A single M4 MacBook does 2.9 TFLOPS. Seven friends with MacBooks match an NVIDIA A100. Alexander Hayes just open-sourced a tool that makes this work over Wi-Fi. It's called AirTrain. Here's how it works: Traditional distributed training (DDP) syncs gradients after every single step. For a 124M parameter model, that's ~500MB exchanged per step. You need 50 GB/s of sustained bandwidth. Impossible over Wi-Fi. AirTrain uses the DiLoCo algorithm. Each Mac trains independently for 500 steps, then syncs only the difference. One sync per 500 steps instead of one per step. 500x less network communication. Wi-Fi actually works. The entire sync takes ~2 seconds. Here's what makes it wild: → Zero-config discovery. Devices find each other automatically via mDNS/Bonjour. Same protocol as AirDrop. → Fault tolerant. Nodes can join and leave mid-training without killing the run. → Checkpoint relay. Train for a few hours, export a checkpoint, hand it off to someone else to continue. Like a relay race for ML training. → Built on Apple's MLX framework. Native to M1/M2/M3/M4/M5 unified memory. No host-to-device copy overhead. → Local dashboard. Real-time loss curves, peer monitoring, throughput metrics in your browser. Here's the wildest part: An M4 Max with 128GB unified memory can train a 70B parameter model without offloading. An NVIDIA RTX 4090 has 24GB VRAM. Apple Silicon gets ~245-460 GFLOPS per watt. Training on MacBooks costs almost nothing in electricity compared to cloud GPUs. And there are hundreds of millions of Apple Silicon Macs in the world. The math: Traditional DDP: 1 sync per step = 50 GB/s required AirTrain (DiLoCo): 1 sync per 500 steps = 0.1 GB/s required Wi-Fi handles 0.1 GB/s. That's it. That's the breakthrough. They even built a community platform at with live session browsing, checkpoint sharing, and a contributor leaderboard. Training a 124M parameter GPT-2? Instead of renting cloud GPUs at $3/hr, pool three MacBooks in a coffee shop and train for free. MIT licensed. Built in Python. 1 contributor. Early stage but the idea is insane. 100% Open Source. (Link in the comments)

Cloud GPU training is a scam. A single M4 MacBook does 2.9 TFLOPS. Seven friends with MacBooks match an NVIDIA A100. Alexander Hayes just open-sourced a tool that makes this work over Wi-Fi. It's called AirTrain. Here's how it works: Traditional distributed training (DDP) syncs gradients after every single step. For a 124M parameter model, that's ~500MB exchanged per step. You need 50 GB/s of sustained bandwidth. Impossible over Wi-Fi. AirTrain uses the DiLoCo algorithm. Each Mac trains independently for 500 steps, then syncs only the difference. One sync per 500 steps instead of one per step. 500x less network communication. Wi-Fi actually works. The entire sync takes ~2 seconds. Here's what makes it wild: → Zero-config discovery. Devices find each other automatically via mDNS/Bonjour. Same protocol as AirDrop. → Fault tolerant. Nodes can join and leave mid-training without killing the run. → Checkpoint relay. Train for a few hours, export a checkpoint, hand it off to someone else to continue. Like a relay race for ML training. → Built on Apple's MLX framework. Native to M1/M2/M3/M4/M5 unified memory. No host-to-device copy overhead. → Local dashboard. Real-time loss curves, peer monitoring, throughput metrics in your browser. Here's the wildest part: An M4 Max with 128GB unified memory can train a 70B parameter model without offloading. An NVIDIA RTX 4090 has 24GB VRAM. Apple Silicon gets ~245-460 GFLOPS per watt. Training on MacBooks costs almost nothing in electricity compared to cloud GPUs. And there are hundreds of millions of Apple Silicon Macs in the world. The math: Traditional DDP: 1 sync per step = 50 GB/s required AirTrain (DiLoCo): 1 sync per 500 steps = 0.1 GB/s required Wi-Fi handles 0.1 GB/s. That's it. That's the breakthrough. They even built a community platform at with live session browsing, checkpoint sharing, and a contributor leaderboard. Training a 124M parameter GPT-2? Instead of renting cloud GPUs at $3/hr, pool three MacBooks in a coffee shop and train for free. MIT licensed. Built in Python. 1 contributor. Early stage but the idea is insane. 100% Open Source. (Link in the comments)

Guri Singh

160,201 views • 3 months ago

I gave a talk at GPU MODE workshop last week on llm.c - the origin story of llm.c - being naked in the world without PyTorch and having to re-invent Array, Autograd, Device, Dtype, Compile, Distributed - how to port a PyTorch layer to 1) explicit PyTorch - and then to 2) write the backward pass - 3) port forward & backward pass to C - 4) string all the layers together - achieving one file of C with no dependencies that compiles and runs ~instantly, where all memory is pre-planned and allocated a single time, fully deterministic, portable code that can run on a potato or a von Neumann probe - how most of llm.c was built at 1am-7am in a water villa porch in Maldives and why this is the recommended way to develop software - convert all of it to run in CUDA on GPU in fp32 - port matmul to cuBLAS - port attention to cuDNN flash-attention - introduce bfloat16 mixed precision - introduce many more optimizations and features like kernel fusions, Packed128, stochastic rounding, full determinism - add multi-GPU training, NCCL, sharded optimizer - add multi-node with MPI or file system or socket - reproduce GPT-2 (1.6B) on one 8XH100 node in 24 hours for $672 in llm.c, achieving (at the time) 29% less memory, 19% faster training that PyTorch nightly, and much faster compile & run - how open source development attracts Avengers from the internet - port to training Llama 3 imminent (branch exists) - many other notable forks - last thought: how software abstractions like Python/PyTorch and everything else really exist only because humans are finite in knowledge, IQ and attention, and how with increasing AI capability LLMs may export custom binaries like llm.c for any application directly, tearing apart and refactoring all abstractions as needed. More links in reply

I gave a talk at GPU MODE workshop last week on llm.c - the origin story of llm.c - being naked in the world without PyTorch and having to re-invent Array, Autograd, Device, Dtype, Compile, Distributed - how to port a PyTorch layer to 1) explicit PyTorch - and then to 2) write the backward pass - 3) port forward & backward pass to C - 4) string all the layers together - achieving one file of C with no dependencies that compiles and runs ~instantly, where all memory is pre-planned and allocated a single time, fully deterministic, portable code that can run on a potato or a von Neumann probe - how most of llm.c was built at 1am-7am in a water villa porch in Maldives and why this is the recommended way to develop software - convert all of it to run in CUDA on GPU in fp32 - port matmul to cuBLAS - port attention to cuDNN flash-attention - introduce bfloat16 mixed precision - introduce many more optimizations and features like kernel fusions, Packed128, stochastic rounding, full determinism - add multi-GPU training, NCCL, sharded optimizer - add multi-node with MPI or file system or socket - reproduce GPT-2 (1.6B) on one 8XH100 node in 24 hours for $672 in llm.c, achieving (at the time) 29% less memory, 19% faster training that PyTorch nightly, and much faster compile & run - how open source development attracts Avengers from the internet - port to training Llama 3 imminent (branch exists) - many other notable forks - last thought: how software abstractions like Python/PyTorch and everything else really exist only because humans are finite in knowledge, IQ and attention, and how with increasing AI capability LLMs may export custom binaries like llm.c for any application directly, tearing apart and refactoring all abstractions as needed. More links in reply

Andrej Karpathy

336,782 views • 1 year ago

Get PAID to use NECT Nectar (NECT), the native Berachain CDP stablecoin powered by PoL is seeing huge growth through integrations, LP’s being set up and explosive usage! Here is a secret most don’t know. You can earn POLLEN by using NECT. Here are a few ways to start earning today. 🟠Set up ANY token LP with NECT, start earning POLLEN. 🟠Integrate NECT into your dApp, start earning POLLEN with usage. 🟠Not sure what stablecoin to hold? Choose the one that pays you. NECT. No need to ask for permission. Just use NECT. How does this work? Every NECT in circulation is earning interest rates inside the Beraborrow Protocol, POLLEN is then distributed to users of NECT. Interest rates are then passed through the POLLEN flywheel that uses PoL to maximise the value distributed back to POLLEN. Allowing for the continued sustainable explosive growth to occur! Resulting in more NECT usage -> more interest rates -> more POLLEN to NECT users. NECT usage is picked up automatically with incentives claimable once POLLEN is live! Post TGE, we have something cooking to make it even easier to get your POLLEN! What truly sets NECT apart is its zero-friction approach: you can start using it today with no governance hoops to jump through or prior approvals required. Incentives are emitted based on TVL, ensuring a seamless user experience for both new and established projects. Get NECT or get REKT. Ooga Matata.

Get PAID to use NECT Nectar (NECT), the native Berachain CDP stablecoin powered by PoL is seeing huge growth through integrations, LP’s being set up and explosive usage! Here is a secret most don’t know. You can earn POLLEN by using NECT. Here are a few ways to start earning today. 🟠Set up ANY token LP with NECT, start earning POLLEN. 🟠Integrate NECT into your dApp, start earning POLLEN with usage. 🟠Not sure what stablecoin to hold? Choose the one that pays you. NECT. No need to ask for permission. Just use NECT. How does this work? Every NECT in circulation is earning interest rates inside the Beraborrow Protocol, POLLEN is then distributed to users of NECT. Interest rates are then passed through the POLLEN flywheel that uses PoL to maximise the value distributed back to POLLEN. Allowing for the continued sustainable explosive growth to occur! Resulting in more NECT usage -> more interest rates -> more POLLEN to NECT users. NECT usage is picked up automatically with incentives claimable once POLLEN is live! Post TGE, we have something cooking to make it even easier to get your POLLEN! What truly sets NECT apart is its zero-friction approach: you can start using it today with no governance hoops to jump through or prior approvals required. Incentives are emitted based on TVL, ensuring a seamless user experience for both new and established projects. Get NECT or get REKT. Ooga Matata.

Beraborrow (🐻⛓️🍯)

21,799 views • 1 year ago

The bottleneck in AI has quietly shifted. - It's not the models. They are capable. - It's not the frameworks. They are mature. - It's not even the data, in many cases. When you want to train a model today, the first question isn't "what architecture should I use?" Instead, it's: "Where am I going to get infrastructure that actually works?" Not just GPUs but the entire stack: compute, deployment, scaling, storage. The traditional path is major cloud providers or specialized GPU clouds. Both have the same problem: they're built for enterprises with committed workloads, minimum spend requirements, contract negotiations, and involve quota approvals that take days. Even the "on-demand" options require you to piece together training, deployment, and scaling across different services. By the time you're actually training, hours, if not days, have passed. And there's a subtler cost: part of your brain is always managing infrastructure instead of thinking about the actual problem. I've been using Runpod for a while now, and it's the closest I've found to infrastructure that just disappears. I pay for the serverless solution by the second, and stop when I'm done. This sounds like it should be the default across all providers, but it isn't. For instance, when I'm prototyping, I don't need an H100. Instead, I need the flexibility to use cheaper GPUs that are actually available, where I can iterate fast and not worry about cost. An A40 at a few cents per hour is perfect for this. Then, when the approach is validated, I scale up. This matches how good engineering actually works. Running distributed training across multiple nodes for multi-GPU training usually requires significant infra work. RunPod abstracts most of this away. A lot of the advantage in AI comes from iteration speed. Infra that adds days of latency to that loop is a real cost, even if it's hard to measure. But good infra gets out of your way. It's available when you need it, invisible when you don't. In the video below, I have shown a simple model training workflow trained using PyTorch in Jupyter Lab. It runs in a dedicated PyTorch Pod hosted on Runpod, and I worked with the team to put this together for you. Find a link to start using Runpod in the replies!

The bottleneck in AI has quietly shifted. - It's not the models. They are capable. - It's not the frameworks. They are mature. - It's not even the data, in many cases. When you want to train a model today, the first question isn't "what architecture should I use?" Instead, it's: "Where am I going to get infrastructure that actually works?" Not just GPUs but the entire stack: compute, deployment, scaling, storage. The traditional path is major cloud providers or specialized GPU clouds. Both have the same problem: they're built for enterprises with committed workloads, minimum spend requirements, contract negotiations, and involve quota approvals that take days. Even the "on-demand" options require you to piece together training, deployment, and scaling across different services. By the time you're actually training, hours, if not days, have passed. And there's a subtler cost: part of your brain is always managing infrastructure instead of thinking about the actual problem. I've been using Runpod for a while now, and it's the closest I've found to infrastructure that just disappears. I pay for the serverless solution by the second, and stop when I'm done. This sounds like it should be the default across all providers, but it isn't. For instance, when I'm prototyping, I don't need an H100. Instead, I need the flexibility to use cheaper GPUs that are actually available, where I can iterate fast and not worry about cost. An A40 at a few cents per hour is perfect for this. Then, when the approach is validated, I scale up. This matches how good engineering actually works. Running distributed training across multiple nodes for multi-GPU training usually requires significant infra work. RunPod abstracts most of this away. A lot of the advantage in AI comes from iteration speed. Infra that adds days of latency to that loop is a real cost, even if it's hard to measure. But good infra gets out of your way. It's available when you need it, invisible when you don't. In the video below, I have shown a simple model training workflow trained using PyTorch in Jupyter Lab. It runs in a dedicated PyTorch Pod hosted on Runpod, and I worked with the team to put this together for you. Find a link to start using Runpod in the replies!

Avi Chawla

13,696 views • 6 months ago

𝐖𝐓𝐅 𝐢𝐬 𝐁𝐥𝐨𝐜𝐤𝐜𝐡𝐚𝐢𝐧 𝐈𝐧𝐝𝐞𝐱𝐢𝐧𝐠?! 🤔 An indexer's job is to process and organize the raw data created on-chain, into a format that's easily accessible and understandable, and then create structured databases that can be queried efficiently by developers so dApps and services can be built and maintained. 🗄️ For example, just think about your favourite web3 marketplace, exchange, game, analytics dashboard etc….the data on those sites is provided by indexers.👌 HOWEVER, on Bitcoin, 99.9% of indexers are centralised, so as it stands, we all have to just trust their version(s) of ‘the truth’! 👀 ....."𝘵𝘳𝘶𝘴𝘵 𝘶𝘴 𝘣𝘳𝘰, 𝘸𝘦 𝘸𝘰𝘯'𝘵 𝘤𝘩𝘢𝘯𝘨𝘦 𝘰𝘳 𝘮𝘢𝘯𝘪𝘱𝘶𝘭𝘢𝘵𝘦 𝘵𝘩𝘦 𝘥𝘢𝘵𝘢, 𝘸𝘦 𝘱𝘳𝘰𝘮𝘪𝘴𝘦!" 😬 At this point, you should pause for a second and ask yourself this one question: 🔽 What was the point of Satoshi creating such a perfectly decentralised data creator like Bitcoin, if we then have to trust one company’s interpretation of that data?! 🙈 This is why 𝐓𝐫𝐚𝐜 𝐂𝐨𝐫𝐞 from @trac_btc is SO vital! 🙏 It is #Bitcoin’s fully decentralised indexer, and it will make on-chain data reliably accessible through its thousands of distributed validator nodes, that MUST first reach consensus, similarly to Bitcoin when it creates the raw data. 🤝 Importantly, those who run validator nodes on their device, and help to support the decentralisation of the Trac network, will be rewarded with token APY! 🪙 Just like for those who run Bitcoin nodes, if you contribute to the decentralisation of the Trac network, you should be rewarded! 🙌 Decentralised indexers are one of the most fundamentally important pieces of infrastructure within a blockchain ecosystem, and they play a crucial role in fostering a scalable and secure ecosystem for blockchain-based economies to thrive and grow! 📈 𝐑𝐞𝐦𝐢𝐧𝐝𝐞𝐫: If you'd like to own/run your own Trac Validator Node, we are currently running regular rounds, with 300 Validator Licences purchasable per round, and with increasing price points. 📈 Follow @trac_btc to stay up to date with the sales rounds in the coming weeks and months 🔔

𝐖𝐓𝐅 𝐢𝐬 𝐁𝐥𝐨𝐜𝐤𝐜𝐡𝐚𝐢𝐧 𝐈𝐧𝐝𝐞𝐱𝐢𝐧𝐠?! 🤔 An indexer's job is to process and organize the raw data created on-chain, into a format that's easily accessible and understandable, and then create structured databases that can be queried efficiently by developers so dApps and services can be built and maintained. 🗄️ For example, just think about your favourite web3 marketplace, exchange, game, analytics dashboard etc….the data on those sites is provided by indexers.👌 HOWEVER, on Bitcoin, 99.9% of indexers are centralised, so as it stands, we all have to just trust their version(s) of ‘the truth’! 👀 ....."𝘵𝘳𝘶𝘴𝘵 𝘶𝘴 𝘣𝘳𝘰, 𝘸𝘦 𝘸𝘰𝘯'𝘵 𝘤𝘩𝘢𝘯𝘨𝘦 𝘰𝘳 𝘮𝘢𝘯𝘪𝘱𝘶𝘭𝘢𝘵𝘦 𝘵𝘩𝘦 𝘥𝘢𝘵𝘢, 𝘸𝘦 𝘱𝘳𝘰𝘮𝘪𝘴𝘦!" 😬 At this point, you should pause for a second and ask yourself this one question: 🔽 What was the point of Satoshi creating such a perfectly decentralised data creator like Bitcoin, if we then have to trust one company’s interpretation of that data?! 🙈 This is why 𝐓𝐫𝐚𝐜 𝐂𝐨𝐫𝐞 from @trac_btc is SO vital! 🙏 It is #Bitcoin’s fully decentralised indexer, and it will make on-chain data reliably accessible through its thousands of distributed validator nodes, that MUST first reach consensus, similarly to Bitcoin when it creates the raw data. 🤝 Importantly, those who run validator nodes on their device, and help to support the decentralisation of the Trac network, will be rewarded with token APY! 🪙 Just like for those who run Bitcoin nodes, if you contribute to the decentralisation of the Trac network, you should be rewarded! 🙌 Decentralised indexers are one of the most fundamentally important pieces of infrastructure within a blockchain ecosystem, and they play a crucial role in fostering a scalable and secure ecosystem for blockchain-based economies to thrive and grow! 📈 𝐑𝐞𝐦𝐢𝐧𝐝𝐞𝐫: If you'd like to own/run your own Trac Validator Node, we are currently running regular rounds, with 300 Validator Licences purchasable per round, and with increasing price points. 📈 Follow @trac_btc to stay up to date with the sales rounds in the coming weeks and months 🔔

Trac Network

56,691 views • 2 years ago

Dario Amodei just revealed that the AI training bottleneck everyone is worried about doesn’t exist anymore. The industry spent years obsessed with scraping the open web. More data. More text. More human output to feed the models. Amodei: “I don’t think data is quite the most central thing anymore.” The shift is fundamental. Amodei: “Static data is becoming less important. A lot of the data we use today is RL environments that we train on. Dynamic data that the model creates itself.” Not scraped. Not licensed. Not written by humans. Generated by the model through pure trial and error. When you train on complex math or agentic coding, you don’t feed it a textbook. You give it an environment. The model experiments. Fails. Adjusts. Tries again. Amodei: “You’re getting some math problems and the model experiments with trying the math problems.” It generates its own experience. Millions of iterations. Each one building on the last. No human required. This destroys the entire narrative around AI hitting a data wall. You cannot throttle a competitor by locking down copyright. Cannot slow the race by putting up a paywall. When a model learns through its own synthetic experience, the open web becomes irrelevant. The only true bottleneck left is compute. And this is where the geopolitical stakes become impossible to overstate. The nation that wins the compute race doesn’t just build smarter models. It builds models that generate their own intelligence, compounding on themselves, iterating past every limit human knowledge ever imposed. We are no longer training AI on the past. We are letting it simulate the future. The machine has stopped reading the dictionary. It’s doing the math itself now.

Dario Amodei just revealed that the AI training bottleneck everyone is worried about doesn’t exist anymore. The industry spent years obsessed with scraping the open web. More data. More text. More human output to feed the models. Amodei: “I don’t think data is quite the most central thing anymore.” The shift is fundamental. Amodei: “Static data is becoming less important. A lot of the data we use today is RL environments that we train on. Dynamic data that the model creates itself.” Not scraped. Not licensed. Not written by humans. Generated by the model through pure trial and error. When you train on complex math or agentic coding, you don’t feed it a textbook. You give it an environment. The model experiments. Fails. Adjusts. Tries again. Amodei: “You’re getting some math problems and the model experiments with trying the math problems.” It generates its own experience. Millions of iterations. Each one building on the last. No human required. This destroys the entire narrative around AI hitting a data wall. You cannot throttle a competitor by locking down copyright. Cannot slow the race by putting up a paywall. When a model learns through its own synthetic experience, the open web becomes irrelevant. The only true bottleneck left is compute. And this is where the geopolitical stakes become impossible to overstate. The nation that wins the compute race doesn’t just build smarter models. It builds models that generate their own intelligence, compounding on themselves, iterating past every limit human knowledge ever imposed. We are no longer training AI on the past. We are letting it simulate the future. The machine has stopped reading the dictionary. It’s doing the math itself now.

Dustin

149,610 views • 5 months ago

🔥 JUST IN: Open-source robotics dataset from 100% real-world scenarios! 🤯 Chinese robotics company AGIBOT just released AGIBOT WORLD 2026, an open-source dataset systematically covering key embodied AI research directions. Built entirely from real-world environments: commercial spaces, and homes. Collected using AGIBOT G2 robots in free-form collection mode, providing structured, accurately annotated, high-quality data. Digital twin technology creates 1:1 scale replicas in simulation matching the real environments. Both real-world and simulation data are open-sourced. The AGIBOT G2 platform collects multiple data types simultaneously: RGB(D) cameras, tactile sensors, force sensors, LiDAR, IMU, and full-body joint states. Whole-body control coordinates arms, waist, and hands for complex tasks. First-person teleoperation lets operators control the robot from its perspective. The tasks covered are fine-grained manipulation, ultra-long-horizon tasks, spatial navigation, dual-arm coordination, and multi-agent/human-robot collaboration. The dataset includes error-recovery trajectories with annotations. Most datasets only show successful demonstrations. AGIBOT includes failures and how the robot recovers, teaching models how to handle mistakes. After collection, data is tested through policy training and real-robot deployment to ensure quality. Then processed through industrial quality control with multiple screening and cleaning rounds. Making it open-source accelerates embodied AI research by giving researchers access to high-quality real-world robot data at scale. 🇨🇳 Learn more here: ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

🔥 JUST IN: Open-source robotics dataset from 100% real-world scenarios! 🤯 Chinese robotics company AGIBOT just released AGIBOT WORLD 2026, an open-source dataset systematically covering key embodied AI research directions. Built entirely from real-world environments: commercial spaces, and homes. Collected using AGIBOT G2 robots in free-form collection mode, providing structured, accurately annotated, high-quality data. Digital twin technology creates 1:1 scale replicas in simulation matching the real environments. Both real-world and simulation data are open-sourced. The AGIBOT G2 platform collects multiple data types simultaneously: RGB(D) cameras, tactile sensors, force sensors, LiDAR, IMU, and full-body joint states. Whole-body control coordinates arms, waist, and hands for complex tasks. First-person teleoperation lets operators control the robot from its perspective. The tasks covered are fine-grained manipulation, ultra-long-horizon tasks, spatial navigation, dual-arm coordination, and multi-agent/human-robot collaboration. The dataset includes error-recovery trajectories with annotations. Most datasets only show successful demonstrations. AGIBOT includes failures and how the robot recovers, teaching models how to handle mistakes. After collection, data is tested through policy training and real-robot deployment to ensure quality. Then processed through industrial quality control with multiple screening and cleaning rounds. Making it open-source accelerates embodied AI research by giving researchers access to high-quality real-world robot data at scale. 🇨🇳 Learn more here: ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

Lukas Ziegler

40,583 views • 3 months ago

The Cost of Intelligence is Heading to Zero | Hyperspace P2P Distributed Cache We present to you our breakthrough cross-domain work across AI, distributed systems, cryptography, game theory to solve the primary structural inefficiency at the heart of AI infrastructure: most inference is redundant. Google has reported that only 15% of daily searches are truly novel. The rest are repeats or close variants. LLM inference inherits this same power-law distribution. Enterprise chatbots see 70-80% of queries fall into a handful of intent categories. System prompts are identical across 100% of requests within an application. The KV attention state for "You are a helpful assistant" has been computed billions of times, on millions of GPUs, identically. And yet every AI lab, every startup, every self-hosted deployment - computes and caches these results independently. There is no shared layer. No global memory. Every provider pays the full compute cost for every query, even when the answer already exists somewhere in the network. This is the problem Hyperspace solves where distributed cache operates at three levels, each catching a different class of redundancy: 1. Response cache Same prompt, same model, same parameters - instant cached response from any node in the network. SHA-256 hash lookup via DHT, with cryptographic cache proofs linking every response to its original inference execution. No trust required. Fetchers re-announce as providers, so popular responses replicate naturally across more nodes. 2. KV prefix cache Same system prompt tokens - skip the most expensive part of inference entirely. Prefill (computing Key-Value attention states) is deterministic: same model plus same tokens always produces identical KV state. The network caches these states using erasure coding and distributes them via the routing network. New questions that share a common prefix resume generation from cached state instead of recomputing from scratch. 3. Routing to cached nodes Instead of transferring KV state across the network for every request, Hyperspace routes the request to the node that already has the state loaded in VRAM. The request goes to the cache, not the cache to the request. Together, these three layers mean that 70-90% of inference requests at network scale never require full GPU computation. This work doesn't exist in isolation. It builds on research from across the industry: SGLang's RadixAttention demonstrated that automatic prefix sharing can yield up to 5x speedup on structured LLM workloads. Moonshot AI's Mooncake built an entire KV-cache-centric disaggregated architecture for production serving at Kimi. Anthropic, OpenAI, and Google all launched prompt caching products in 2024 - priced at 50-90% discounts - because system prompt reuse is so pervasive that it changes the economics of inference. What all of these systems share is a common limitation: they operate within a single organization's infrastructure. SGLang caches prefixes within one server. Mooncake disaggregates KV cache within one datacenter. Anthropic's prompt caching works within one API provider's fleet. None of them can share cached state across organizational boundaries. Hyperspace removes this boundary. The cache is global. A response computed by a node in Tokyo is immediately available to a node in Berlin. A KV prefix state generated for Qwen-32B on one machine is verifiable and reusable by any other machine running the same model. The routing network provides the delivery guarantees, the erasure coding provides the redundancy, and the cache proofs provide the trust. What this means for the cost of intelligence Big AI labs scale linearly: twice the users means twice the GPU spend. Every query is a cost center. Their internal caching helps, but it's siloed - Lab A's cache can't serve Lab B's users, and neither can serve a self-hosted Llama deployment. Hyperspace scales sub-linearly. Every new node that joins the network adds to the global cache. Every inference result enriches the cache for all future requests. The cache hit rate rises with network size because query distributions follow a power law - the most common questions are asked exponentially more often than rare ones. The implication is simple: as the network grows, the effective cost per inference drops. Not linearly. Logarithmically. At 10 million nodes, we estimate 75-90% of all inference requests can be served from cache, eliminating 400,000+ MWh of energy consumption per year and avoiding over 200,000 tons of CO2 emissions. The first person to ask a question pays the compute cost. Everyone after them gets the answer for free, with cryptographic proof that it's authentic. Training is competitive. Inference is shared Open-weight models are converging on quality with closed models. Labs will continue to differentiate on training - data curation, architecture innovation, RLHF tuning. That's where the real intellectual property lives. But inference is a commodity. Two copies of Qwen-32B running the same prompt produce the same KV state and the same response, byte for byte, regardless of whose GPU runs the matrix multiplication. There is no moat in multiplying matrices. The moat is in training the weights. A global distributed cache makes this separation explicit. It doesn't matter who trained the model. Once the weights are open, the inference cost approaches zero at scale - because the network remembers every answer and can prove it's correct. No lab, no matter how well-funded, can match this. They cannot share caches across competitors. They scale linearly. The network scales logarithmically. The marginal cost of intelligence approaches zero. That's the endgame.

The Cost of Intelligence is Heading to Zero | Hyperspace P2P Distributed Cache We present to you our breakthrough cross-domain work across AI, distributed systems, cryptography, game theory to solve the primary structural inefficiency at the heart of AI infrastructure: most inference is redundant. Google has reported that only 15% of daily searches are truly novel. The rest are repeats or close variants. LLM inference inherits this same power-law distribution. Enterprise chatbots see 70-80% of queries fall into a handful of intent categories. System prompts are identical across 100% of requests within an application. The KV attention state for "You are a helpful assistant" has been computed billions of times, on millions of GPUs, identically. And yet every AI lab, every startup, every self-hosted deployment - computes and caches these results independently. There is no shared layer. No global memory. Every provider pays the full compute cost for every query, even when the answer already exists somewhere in the network. This is the problem Hyperspace solves where distributed cache operates at three levels, each catching a different class of redundancy: 1. Response cache Same prompt, same model, same parameters - instant cached response from any node in the network. SHA-256 hash lookup via DHT, with cryptographic cache proofs linking every response to its original inference execution. No trust required. Fetchers re-announce as providers, so popular responses replicate naturally across more nodes. 2. KV prefix cache Same system prompt tokens - skip the most expensive part of inference entirely. Prefill (computing Key-Value attention states) is deterministic: same model plus same tokens always produces identical KV state. The network caches these states using erasure coding and distributes them via the routing network. New questions that share a common prefix resume generation from cached state instead of recomputing from scratch. 3. Routing to cached nodes Instead of transferring KV state across the network for every request, Hyperspace routes the request to the node that already has the state loaded in VRAM. The request goes to the cache, not the cache to the request. Together, these three layers mean that 70-90% of inference requests at network scale never require full GPU computation. This work doesn't exist in isolation. It builds on research from across the industry: SGLang's RadixAttention demonstrated that automatic prefix sharing can yield up to 5x speedup on structured LLM workloads. Moonshot AI's Mooncake built an entire KV-cache-centric disaggregated architecture for production serving at Kimi. Anthropic, OpenAI, and Google all launched prompt caching products in 2024 - priced at 50-90% discounts - because system prompt reuse is so pervasive that it changes the economics of inference. What all of these systems share is a common limitation: they operate within a single organization's infrastructure. SGLang caches prefixes within one server. Mooncake disaggregates KV cache within one datacenter. Anthropic's prompt caching works within one API provider's fleet. None of them can share cached state across organizational boundaries. Hyperspace removes this boundary. The cache is global. A response computed by a node in Tokyo is immediately available to a node in Berlin. A KV prefix state generated for Qwen-32B on one machine is verifiable and reusable by any other machine running the same model. The routing network provides the delivery guarantees, the erasure coding provides the redundancy, and the cache proofs provide the trust. What this means for the cost of intelligence Big AI labs scale linearly: twice the users means twice the GPU spend. Every query is a cost center. Their internal caching helps, but it's siloed - Lab A's cache can't serve Lab B's users, and neither can serve a self-hosted Llama deployment. Hyperspace scales sub-linearly. Every new node that joins the network adds to the global cache. Every inference result enriches the cache for all future requests. The cache hit rate rises with network size because query distributions follow a power law - the most common questions are asked exponentially more often than rare ones. The implication is simple: as the network grows, the effective cost per inference drops. Not linearly. Logarithmically. At 10 million nodes, we estimate 75-90% of all inference requests can be served from cache, eliminating 400,000+ MWh of energy consumption per year and avoiding over 200,000 tons of CO2 emissions. The first person to ask a question pays the compute cost. Everyone after them gets the answer for free, with cryptographic proof that it's authentic. Training is competitive. Inference is shared Open-weight models are converging on quality with closed models. Labs will continue to differentiate on training - data curation, architecture innovation, RLHF tuning. That's where the real intellectual property lives. But inference is a commodity. Two copies of Qwen-32B running the same prompt produce the same KV state and the same response, byte for byte, regardless of whose GPU runs the matrix multiplication. There is no moat in multiplying matrices. The moat is in training the weights. A global distributed cache makes this separation explicit. It doesn't matter who trained the model. Once the weights are open, the inference cost approaches zero at scale - because the network remembers every answer and can prove it's correct. No lab, no matter how well-funded, can match this. They cannot share caches across competitors. They scale linearly. The network scales logarithmically. The marginal cost of intelligence approaches zero. That's the endgame.

Varun

37,542 views • 4 months ago

I trained a 100 million parameter DeepSeek V3 LLM from scratch Here's what you need to know. Previously I trained traditional GPT-2 architecture which has become obsolete with recent LLM advancements. Most recent models like Llama, Mistral, DeepSeek, and GPT-4 use latest architectures. ✦ Model Configuration of my SLM DeepSeek V3 - Parameters: 109,032,032 - Embedding Dimension: 512 - Layers: 8 - Heads: 8 - Experts (MoE): 8 - Experts per token: 2 ✦ DeepSeek brings major architectural changes: - Multi Head Latent Attention - Mixture of Experts - RMS Norm - Multi Token Prediction ✦ Dataset Challenge - TinyStories is great for learning SLMs. I trained GPT-2 on it previously with good results. - But I needed a more challenging dataset. - If I use TinyStories again on DeepSeek, how would I know MHLA, MoE or MTP works better than old architecture? - The old architecture can handle it, so new DeepSeek would too without utilizing latest advancements. That's why I moved to FineWeb-Edu dataset Thanks Yuvraj Singh (smolhub.com) for the suggestion for this dataset ✦ Training Journey - Rented A100 PCIe GPU and trained the model. - Did test runs. During final run, model was 65% trained but stopped due to glitch after 4 hours. - Fixed all edge cases and ran training again with increased config parameters. - Final training: 7 hours, 20,000 epochs 𝐓𝐨𝐭𝐚𝐥 𝐆𝐏𝐔 𝐜𝐨𝐬𝐭: $17 - $9.53 for main 7-hour run - $7.42 for experiments and demos ✦ Reflection Amazing long project that taught me latest architectural advancements. I'll reimplement and revisit after a few weeks because there's too much complexity, mostly in Multi Head Latent Attention part. Need to make concepts stronger. Code Final trained Model Dataset Resources Huge shoutout to Raj Dandekar again for creating one of the most detailed video series about DeepSeek - this was my primary resource for the implementation. Playlist Blogs by Maarten Grootendorst These are excellent visual blogs to understand MoE in detail. Thanks Maarten for your amazing contributions to the community through your books and blogs Blogs on MoE Implemention of MoE from scratch by @aviTwit3 One of the most detailed blogs on implementing Mixture of Experts. Thanks Avinash for this blog - it helped me understand Mixture of Experts much better. If you're someone in the 𝐌𝐋 & 𝐋𝐋𝐌 space, would love to 𝐜𝐨𝐧𝐧𝐞𝐜𝐭 and discuss this field in general, so give a follow up for that.

I trained a 100 million parameter DeepSeek V3 LLM from scratch Here's what you need to know. Previously I trained traditional GPT-2 architecture which has become obsolete with recent LLM advancements. Most recent models like Llama, Mistral, DeepSeek, and GPT-4 use latest architectures. ✦ Model Configuration of my SLM DeepSeek V3 - Parameters: 109,032,032 - Embedding Dimension: 512 - Layers: 8 - Heads: 8 - Experts (MoE): 8 - Experts per token: 2 ✦ DeepSeek brings major architectural changes: - Multi Head Latent Attention - Mixture of Experts - RMS Norm - Multi Token Prediction ✦ Dataset Challenge - TinyStories is great for learning SLMs. I trained GPT-2 on it previously with good results. - But I needed a more challenging dataset. - If I use TinyStories again on DeepSeek, how would I know MHLA, MoE or MTP works better than old architecture? - The old architecture can handle it, so new DeepSeek would too without utilizing latest advancements. That's why I moved to FineWeb-Edu dataset Thanks Yuvraj Singh (smolhub.com) for the suggestion for this dataset ✦ Training Journey - Rented A100 PCIe GPU and trained the model. - Did test runs. During final run, model was 65% trained but stopped due to glitch after 4 hours. - Fixed all edge cases and ran training again with increased config parameters. - Final training: 7 hours, 20,000 epochs 𝐓𝐨𝐭𝐚𝐥 𝐆𝐏𝐔 𝐜𝐨𝐬𝐭: $17 - $9.53 for main 7-hour run - $7.42 for experiments and demos ✦ Reflection Amazing long project that taught me latest architectural advancements. I'll reimplement and revisit after a few weeks because there's too much complexity, mostly in Multi Head Latent Attention part. Need to make concepts stronger. Code Final trained Model Dataset Resources Huge shoutout to Raj Dandekar again for creating one of the most detailed video series about DeepSeek - this was my primary resource for the implementation. Playlist Blogs by Maarten Grootendorst These are excellent visual blogs to understand MoE in detail. Thanks Maarten for your amazing contributions to the community through your books and blogs Blogs on MoE Implemention of MoE from scratch by @aviTwit3 One of the most detailed blogs on implementing Mixture of Experts. Thanks Avinash for this blog - it helped me understand Mixture of Experts much better. If you're someone in the 𝐌𝐋 & 𝐋𝐋𝐌 space, would love to 𝐜𝐨𝐧𝐧𝐞𝐜𝐭 and discuss this field in general, so give a follow up for that.

Mayank Pratap Singh

48,063 views • 1 year ago

Video: World’s first humanoid robot labor that swaps its own batteries to work endlessly | Jijo Malayil, Interesting Engineering Walker S2 uses dual-battery balancing and standardized modules to boost efficiency and ensure uninterrupted, optimized performance. In a leap for robotics, China’s UBTech has unveiled the Walker S2, the world’s first humanoid robot capable of fully autonomous battery swapping. Designed for non-stop industrial operations, the Walker S2 can replace its own power pack in just three minutes—no human intervention required. Equipped with advanced anthropomorphic bipedal locomotion and a hot-swappable battery system, Walker S2 is built to operate 24/7 across dynamic industrial environments. According to UBTech, the next-generation humanoid robot marks a major milestone in automation, bringing continuous, hands-free performance to the factory floor. In May 2025, UBTech Robotics and Huawei Technologies inked a significant partnership to accelerate the adoption of humanoid robots across China’s factories and households. Uninterrupted robot operations A video posted by the robotics firm opens with the sleek UBTech Walker S2 humanoid robot working in an industrial setting. The highlight, however, is its autonomous battery swap. Walker S2 approaches the charging station, carefully detaches its depleted power pack, and seamlessly installs a fresh one—all within about three minutes—without any human assistance, according to CGTN. The camera captures close-ups of the robot’s articulated limbs and the intelligent battery-handling mechanism, conveying precision and reliability. As the swap completes, Walker S2 resumes its duties, reinforcing the promise of uninterrupted, 24/7 operations in dynamic factory environments. UBTech’s Walker S2 humanoid robot is equipped with advanced dual-battery power balancing technology and uses standardized battery modules to optimize performance, reports CNEVPOST. This dual-battery system allows the robot to automatically switch to a backup battery in case of a main battery failure, ensuring that critical tasks are carried out without interruption. In addition to battery swapping, the robot can intelligently choose between charging and swapping based on task urgency, allowing it to manage energy dynamically and adapt to real-time operational demands. UBTech highlights these features as a step forward in deploying humanoid robots for industrial and domestic applications, combining flexibility, reliability, and autonomy in one intelligent platform. Factory intelligence upgrade Earlier in the year, UBTech unveiled a major advancement in humanoid robot collaboration, claiming the world’s first deployment of multiple humanoids working together across varied industrial tasks. Demonstrated at Zeekr’s 5G-enabled smart factory, the breakthrough centers on UBTech’s “BrainNet” framework, which orchestrates cooperative behavior through a cloud-device intelligence system. BrainNet integrates a “super brain” for high-level decision-making with an “intelligent sub-brain” for distributed multi-robot control. The super brain, powered by a proprietary large-scale multimodal reasoning model, handles complex production-line scheduling and decision-making. Meanwhile, the sub-brain coordinates real-time tasks using cross-field perception and Transformer-based control for dynamic adaptability. Together, they enable the Walker S1 humanoid robots to move beyond isolated operations and perform coordinated tasks with high precision and speed. The system is built on DeepSeek-R1 reasoning technology and trained on real-world data from automotive factory settings. Leveraging Retrieval-Augmented Generation (RAG), the model adapts to specific job functions and improves scalability across workstations. At Zeekr’s facility, dozens of Walker S1s now collaborate on tasks like assembly, inspection, and part handling. Using semantic VSLAM and shared mapping, they coordinate seamlessly via vision-based navigation and agile manipulation. UBTech says this marks a transition to “Practical Training 2.0,” where humanoid robots operate as a swarm, maximizing efficiency and setting the stage for next-generation intelligent manufacturing.

Video: World’s first humanoid robot labor that swaps its own batteries to work endlessly | Jijo Malayil, Interesting Engineering Walker S2 uses dual-battery balancing and standardized modules to boost efficiency and ensure uninterrupted, optimized performance. In a leap for robotics, China’s UBTech has unveiled the Walker S2, the world’s first humanoid robot capable of fully autonomous battery swapping. Designed for non-stop industrial operations, the Walker S2 can replace its own power pack in just three minutes—no human intervention required. Equipped with advanced anthropomorphic bipedal locomotion and a hot-swappable battery system, Walker S2 is built to operate 24/7 across dynamic industrial environments. According to UBTech, the next-generation humanoid robot marks a major milestone in automation, bringing continuous, hands-free performance to the factory floor. In May 2025, UBTech Robotics and Huawei Technologies inked a significant partnership to accelerate the adoption of humanoid robots across China’s factories and households. Uninterrupted robot operations A video posted by the robotics firm opens with the sleek UBTech Walker S2 humanoid robot working in an industrial setting. The highlight, however, is its autonomous battery swap. Walker S2 approaches the charging station, carefully detaches its depleted power pack, and seamlessly installs a fresh one—all within about three minutes—without any human assistance, according to CGTN. The camera captures close-ups of the robot’s articulated limbs and the intelligent battery-handling mechanism, conveying precision and reliability. As the swap completes, Walker S2 resumes its duties, reinforcing the promise of uninterrupted, 24/7 operations in dynamic factory environments. UBTech’s Walker S2 humanoid robot is equipped with advanced dual-battery power balancing technology and uses standardized battery modules to optimize performance, reports CNEVPOST. This dual-battery system allows the robot to automatically switch to a backup battery in case of a main battery failure, ensuring that critical tasks are carried out without interruption. In addition to battery swapping, the robot can intelligently choose between charging and swapping based on task urgency, allowing it to manage energy dynamically and adapt to real-time operational demands. UBTech highlights these features as a step forward in deploying humanoid robots for industrial and domestic applications, combining flexibility, reliability, and autonomy in one intelligent platform. Factory intelligence upgrade Earlier in the year, UBTech unveiled a major advancement in humanoid robot collaboration, claiming the world’s first deployment of multiple humanoids working together across varied industrial tasks. Demonstrated at Zeekr’s 5G-enabled smart factory, the breakthrough centers on UBTech’s “BrainNet” framework, which orchestrates cooperative behavior through a cloud-device intelligence system. BrainNet integrates a “super brain” for high-level decision-making with an “intelligent sub-brain” for distributed multi-robot control. The super brain, powered by a proprietary large-scale multimodal reasoning model, handles complex production-line scheduling and decision-making. Meanwhile, the sub-brain coordinates real-time tasks using cross-field perception and Transformer-based control for dynamic adaptability. Together, they enable the Walker S1 humanoid robots to move beyond isolated operations and perform coordinated tasks with high precision and speed. The system is built on DeepSeek-R1 reasoning technology and trained on real-world data from automotive factory settings. Leveraging Retrieval-Augmented Generation (RAG), the model adapts to specific job functions and improves scalability across workstations. At Zeekr’s facility, dozens of Walker S1s now collaborate on tasks like assembly, inspection, and part handling. Using semantic VSLAM and shared mapping, they coordinate seamlessly via vision-based navigation and agile manipulation. UBTech says this marks a transition to “Practical Training 2.0,” where humanoid robots operate as a swarm, maximizing efficiency and setting the stage for next-generation intelligent manufacturing.

Owen Gregorian

35,637 views • 1 year ago

This is big... Anthropic just announced a model so powerful they won't release it to the public out of fear over the damage it will cause 😨 Claude Mythos Preview found thousands of zero-day exploits in every major operating system and web browser... The numbers are hard to believe: > $50 to find a 27-year-old bug in OpenBSD, one of the most security-hardened operating systems ever built > Under $1,000 to find AND build a fully working remote code execution exploit on FreeBSD that grants unauthenticated root access from anywhere on the internet > Under $2,000 to chain together multiple Linux kernel vulnerabilities into a complete privilege escalation exploit For context: these are the kinds of findings that previously required elite security researchers working for weeks. Anthropic engineers with no formal security training asked Mythos to find exploits overnight. They woke up to working code the next morning. The results were so impressive Anthropic assembled Apple, Google, Microsoft, Amazon, NVIDIA, and seven other organizations into Project Glasswing: A $100M defensive coalition. They're not releasing this model publicly. Instead, they're racing to patch the world's infrastructure before models like this proliferate.

This is big... Anthropic just announced a model so powerful they won't release it to the public out of fear over the damage it will cause 😨 Claude Mythos Preview found thousands of zero-day exploits in every major operating system and web browser... The numbers are hard to believe: > $50 to find a 27-year-old bug in OpenBSD, one of the most security-hardened operating systems ever built > Under $1,000 to find AND build a fully working remote code execution exploit on FreeBSD that grants unauthenticated root access from anywhere on the internet > Under $2,000 to chain together multiple Linux kernel vulnerabilities into a complete privilege escalation exploit For context: these are the kinds of findings that previously required elite security researchers working for weeks. Anthropic engineers with no formal security training asked Mythos to find exploits overnight. They woke up to working code the next morning. The results were so impressive Anthropic assembled Apple, Google, Microsoft, Amazon, NVIDIA, and seven other organizations into Project Glasswing: A $100M defensive coalition. They're not releasing this model publicly. Instead, they're racing to patch the world's infrastructure before models like this proliferate.

Josh Kale

4,063,354 views • 3 months ago

Google just proved that bigger isn't always better. Their 308M parameter model is outperforming models 2x its size. Google just released 𝗘𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝗚𝗲𝗺𝗺𝗮, and it's proving that lightweight embedding models can punch way above their weight class. At just 308M parameters (578MB), it's the new state-of-the-art for models under 500M parameters across MTEB multilingual, English, and code benchmarks. But the really impressive part is that it ranks 8th overall on MTEB(Multilingual, v2) - that's 𝟭𝟳 𝗽𝗹𝗮𝗰𝗲𝘀 above the second-best sub-500M model, and it's delivering performance 𝗰𝗼𝗺𝗽𝗮𝗿𝗮𝗯𝗹𝗲 𝘁𝗼 𝗺𝗼𝗱𝗲𝗹𝘀 𝗻𝗲𝗮𝗿𝗹𝘆 𝗱𝗼𝘂𝗯𝗹𝗲 𝗶𝘁𝘀 𝘀𝗶𝘇𝗲. There are three key parts of their training recipe that sets it apart: 𝟭. 𝗘𝗻𝗰𝗼𝗱𝗲𝗿-𝗗𝗲𝗰𝗼𝗱𝗲𝗿 𝗜𝗻𝗶𝘁𝗶𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻 Instead of starting from a decoder-only Gemma 3 model, they first adapted it to encoder-decoder, then used just the encoder. By basing EmbeddingGemma off an LLM that already has world and language understanding, it gives it a stronger starting point. 𝟮. 𝗧𝗵𝗿𝗲𝗲-𝗟𝗼𝘀𝘀 𝗧𝗿𝗮𝗶𝗻𝗶𝗻𝗴 They combine three different loss functions, instead of just having one: • Contrastive loss (NCE) with in-batch negatives and hardness weighting • Spread-out regularization to ensure embeddings utilize the full space (for quantization and ANN retrieval) • Embedding matching distillation from Gemini Embedding - not just learning from relevance scores, but directly aligning the embedding space with the teacher model 𝟯. 𝗠𝗼𝗱𝗲𝗹 𝗦𝗼𝘂𝗽𝗶𝗻𝗴 Rather than just averaging checkpoints from the same training run, they use optimization techniques to find multiple specialized training mixtures. Each mixture creates an "expert" model in different domains, and averaging all their parameters creates a final model that's actually better than individual models. Extras: • Matryoshka embeddings supporting 768, 512, 256, and 128 dimensions • Quantization-aware training - maintains quality even at int4 precision • 100+ languages from Gemma 3 pretraining • Exceptional performance on low-resource languages (check their XTREME-UP results) Is it the absolute best embedding model? No - Gemini Embedding still leads overall. But that's not really the point. EmbeddingGemma proves you can achieve state-of-the-art performance in a small package that's actually deployable on-device, in low-latency applications, and in resource-constrained environments. This makes good embeddings accessible for use cases that I'm seeing more and more: offline applications, privacy-sensitive deployments, and high-throughput scenarios where inference cost actually matters. Full paper: Shoutout to the EmbeddingGemma team at Google DeepMind for this awesome open source work 💙 and to Daniel Williams for helping me with this video! 🫶

Google just proved that bigger isn't always better. Their 308M parameter model is outperforming models 2x its size. Google just released 𝗘𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝗚𝗲𝗺𝗺𝗮, and it's proving that lightweight embedding models can punch way above their weight class. At just 308M parameters (578MB), it's the new state-of-the-art for models under 500M parameters across MTEB multilingual, English, and code benchmarks. But the really impressive part is that it ranks 8th overall on MTEB(Multilingual, v2) - that's 𝟭𝟳 𝗽𝗹𝗮𝗰𝗲𝘀 above the second-best sub-500M model, and it's delivering performance 𝗰𝗼𝗺𝗽𝗮𝗿𝗮𝗯𝗹𝗲 𝘁𝗼 𝗺𝗼𝗱𝗲𝗹𝘀 𝗻𝗲𝗮𝗿𝗹𝘆 𝗱𝗼𝘂𝗯𝗹𝗲 𝗶𝘁𝘀 𝘀𝗶𝘇𝗲. There are three key parts of their training recipe that sets it apart: 𝟭. 𝗘𝗻𝗰𝗼𝗱𝗲𝗿-𝗗𝗲𝗰𝗼𝗱𝗲𝗿 𝗜𝗻𝗶𝘁𝗶𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻 Instead of starting from a decoder-only Gemma 3 model, they first adapted it to encoder-decoder, then used just the encoder. By basing EmbeddingGemma off an LLM that already has world and language understanding, it gives it a stronger starting point. 𝟮. 𝗧𝗵𝗿𝗲𝗲-𝗟𝗼𝘀𝘀 𝗧𝗿𝗮𝗶𝗻𝗶𝗻𝗴 They combine three different loss functions, instead of just having one: • Contrastive loss (NCE) with in-batch negatives and hardness weighting • Spread-out regularization to ensure embeddings utilize the full space (for quantization and ANN retrieval) • Embedding matching distillation from Gemini Embedding - not just learning from relevance scores, but directly aligning the embedding space with the teacher model 𝟯. 𝗠𝗼𝗱𝗲𝗹 𝗦𝗼𝘂𝗽𝗶𝗻𝗴 Rather than just averaging checkpoints from the same training run, they use optimization techniques to find multiple specialized training mixtures. Each mixture creates an "expert" model in different domains, and averaging all their parameters creates a final model that's actually better than individual models. Extras: • Matryoshka embeddings supporting 768, 512, 256, and 128 dimensions • Quantization-aware training - maintains quality even at int4 precision • 100+ languages from Gemma 3 pretraining • Exceptional performance on low-resource languages (check their XTREME-UP results) Is it the absolute best embedding model? No - Gemini Embedding still leads overall. But that's not really the point. EmbeddingGemma proves you can achieve state-of-the-art performance in a small package that's actually deployable on-device, in low-latency applications, and in resource-constrained environments. This makes good embeddings accessible for use cases that I'm seeing more and more: offline applications, privacy-sensitive deployments, and high-throughput scenarios where inference cost actually matters. Full paper: Shoutout to the EmbeddingGemma team at Google DeepMind for this awesome open source work 💙 and to Daniel Williams for helping me with this video! 🫶

Victoria Slocum

21,592 views • 8 months ago

Google just made every $50K master's degree look like a scam. They dropped "Google Skills" - 3,000+ AI courses from DeepMind, Cloud, and Google Education in one platform. And it's 100% FREE for Google Cloud users. The same content universities charge $60K for: - DeepMind's actual AI research training - 700+ hands-on labs with real cloud environments - Gemini Code Assist built INTO the learning - Direct hiring paths at 150+ companies While everyone's drowning in student debt, smart people are getting: ✓ Skills that actually get you hired ✓ Certificates employers recognize (82% hiring preference) ✓ Zero cost if you have Google Cloud ✓ Or $29/month vs $1,600/month for Udacity The kicker? 26 million people completed courses BEFORE this consolidation. You're competing against people learning AI from the team that BUILT Gemini. How to actually use this (not just browse): 1. Start with "AI Essentials" - no coding required 2. Use the hands-on labs (this is where 90% quit) 3. Get skill badges - they show up on LinkedIn 4. Target Google Cloud certification - top 2 highest paying IT certs 5. Join the 150-company hiring consortium The education industrial complex is panicking because anyone can now: → Learn from DeepMind researchers directly → Practice with $500 in free Cloud credits → Get hired without a degree One person's $60K tuition = 2,070 months of Google Skills. Let that sink in. Comment "SKILLS" and I'll send you: ✓ The exact learning path that gets you hired fastest ✓ Which certifications actually pay ✓ How to access everything free Your competition is still applying to universities. Time to eat their lunch.

Google just made every $50K master's degree look like a scam. They dropped "Google Skills" - 3,000+ AI courses from DeepMind, Cloud, and Google Education in one platform. And it's 100% FREE for Google Cloud users. The same content universities charge $60K for: - DeepMind's actual AI research training - 700+ hands-on labs with real cloud environments - Gemini Code Assist built INTO the learning - Direct hiring paths at 150+ companies While everyone's drowning in student debt, smart people are getting: ✓ Skills that actually get you hired ✓ Certificates employers recognize (82% hiring preference) ✓ Zero cost if you have Google Cloud ✓ Or $29/month vs $1,600/month for Udacity The kicker? 26 million people completed courses BEFORE this consolidation. You're competing against people learning AI from the team that BUILT Gemini. How to actually use this (not just browse): 1. Start with "AI Essentials" - no coding required 2. Use the hands-on labs (this is where 90% quit) 3. Get skill badges - they show up on LinkedIn 4. Target Google Cloud certification - top 2 highest paying IT certs 5. Join the 150-company hiring consortium The education industrial complex is panicking because anyone can now: → Learn from DeepMind researchers directly → Practice with $500 in free Cloud credits → Get hired without a degree One person's $60K tuition = 2,070 months of Google Skills. Let that sink in. Comment "SKILLS" and I'll send you: ✓ The exact learning path that gets you hired fastest ✓ Which certifications actually pay ✓ How to access everything free Your competition is still applying to universities. Time to eat their lunch.

Nozz

568,924 views • 9 months ago

Everyone's building AI agents that run on someone else's server, store memory in someone else's database, and can be shut down by someone else's terms of service. I built one that can't be. FlowClaw is an AI agent that runs on a decentralized distributed computer. Your agent, your conversations, your memory, your tools — all stored onchain on Flow, a distributed network of validator nodes across the world. Not a centralized cloud. Not someone's S3 bucket. A blockchain that functions as censorship-resistant compute and storage for your AI. This isn't a wrapper. Your agent is a Resource — a first-class programmable object in Cadence (Flow's smart contract language) that physically lives in your account's on-chain storage. It can't be duplicated, seized, or deleted by anyone except you. Your encrypted messages, your cognitive memory, your scheduled tasks — they persist on a global distributed ledger that no single entity controls. It's an alpha build. It will break. But it works today on mainnet and I want people to push it this weekend. What it does: You go to authenticate with a passkey (Face ID, Touch ID), and you have a blockchain account in seconds. No wallet. No seed phrase. No tokens needed — gas is sponsored. You're immediately chatting with an AI agent that has real tool execution: live web data, token prices, on-chain balances, Cadence script execution, FLOW transfers. Every message is encrypted client-side before it touches the chain. The agent has a cognitive memory system — it doesn't just remember your last message, it builds molecular memory clusters where related knowledge bonds together for contextual retrieval across sessions. You can spawn sub-agents from a visual canvas to run parallel research. The memory tab shows you exactly what your agent knows. Everything is transparent and everything is yours. 11 smart contracts. No external dependencies. No keeper networks. No account abstraction hacks. Here's the part that matters for the censorship-resistance crowd: FlowClaw supports BYOK — bring your own key. You can plug in any LLM provider. But pair it with Venice and you get the full stack: a censorship-resistant AI model running inference with no content filtering, connected to an agent whose state lives on a decentralized network that no company can shut down, with end-to-end encrypted conversations that nobody can read — not the relay operator, not the LLM provider, not the blockchain validators. Venice doesn't log prompts. Flow can't read your encrypted storage. The relay never sees your plaintext. That's not a privacy policy. That's architecture. You can also use OpenAI, Anthropic, or any OpenAI-compatible provider. The agent platform doesn't care — it's model-agnostic. But the Venice pairing is the one that closes every gap in the stack. For the people tinkering with OpenClaw and the broader open-source agent ecosystem — FlowClaw is exploring what happens when you take the agent off the cloud entirely. Not just open-sourcing the code (though it is), but putting the actual runtime state on a distributed computer. Your agent's memory isn't in a SQLite file on your laptop or a Pinecone index on someone's cluster. It's on-chain, encrypted, and replicated across every validator node on Flow. You own it the way you own a private key — mathematically, not contractually. The blockchain here isn't a gimmick bolted onto an agent for token speculation. It's functioning as the infrastructure layer that replaces AWS. Flow accounts are programmable containers with their own storage, keys, and security capabilities. Passkey authentication works natively because Flow supports P-256 keys at the protocol level — the same curve your phone uses for biometrics. Gas sponsorship works natively because Flow transactions have separate proposer, authorizer, and payer roles built into the protocol. No proxy contracts. No relayers. No ERC-4337. Now here's the part that interests me economically. Every FlowClaw interaction is an on-chain transaction. Every message stored, every memory committed, every session created, every sub-agent spawned. An active user might generate dozens of transactions in a single conversation. Scale that and FlowClaw becomes a real contributor to Flow's transaction volume. Flow.com becomes deflationary at 250 TPS. Applications like FlowClaw that generate high-frequency, storage-heavy transactions are exactly what moves the needle. Every encrypted message uses account storage, which requires FLOW balance to back it. Every transaction burns fees. The more agents running, the more demand for $FLOW — not because of a tokenomics gimmick, but because the protocol literally requires it for compute and storage. FlowClaw doesn't have its own token. The token is $FLOW. The entire platform runs natively on the network — using Flow storage, paying Flow transaction fees, backed by Flow account balances. If FlowClaw succeeds, FLOW captures that value directly. I'm sharing this early because the AI agent space is moving fast and I think the decentralized infrastructure angle is underexplored. Most "crypto AI" projects are tokens with a chatbot attached. FlowClaw is the opposite — it's an agent platform that happens to use a blockchain because the blockchain solves real engineering problems that centralized infrastructure can't. Try it: Github: Create an agent, ask it something, spawn a sub-agent, check your memory tab, pair it with Venice for the full censorship-resistant stack. Break it and tell me what broke. If you think this direction matters, the best thing you can do is use it and give feedback. Your AI agent should be yours. Not your provider's. Not your platform's. Yours.

Everyone's building AI agents that run on someone else's server, store memory in someone else's database, and can be shut down by someone else's terms of service. I built one that can't be. FlowClaw is an AI agent that runs on a decentralized distributed computer. Your agent, your conversations, your memory, your tools — all stored onchain on Flow, a distributed network of validator nodes across the world. Not a centralized cloud. Not someone's S3 bucket. A blockchain that functions as censorship-resistant compute and storage for your AI. This isn't a wrapper. Your agent is a Resource — a first-class programmable object in Cadence (Flow's smart contract language) that physically lives in your account's on-chain storage. It can't be duplicated, seized, or deleted by anyone except you. Your encrypted messages, your cognitive memory, your scheduled tasks — they persist on a global distributed ledger that no single entity controls. It's an alpha build. It will break. But it works today on mainnet and I want people to push it this weekend. What it does: You go to authenticate with a passkey (Face ID, Touch ID), and you have a blockchain account in seconds. No wallet. No seed phrase. No tokens needed — gas is sponsored. You're immediately chatting with an AI agent that has real tool execution: live web data, token prices, on-chain balances, Cadence script execution, FLOW transfers. Every message is encrypted client-side before it touches the chain. The agent has a cognitive memory system — it doesn't just remember your last message, it builds molecular memory clusters where related knowledge bonds together for contextual retrieval across sessions. You can spawn sub-agents from a visual canvas to run parallel research. The memory tab shows you exactly what your agent knows. Everything is transparent and everything is yours. 11 smart contracts. No external dependencies. No keeper networks. No account abstraction hacks. Here's the part that matters for the censorship-resistance crowd: FlowClaw supports BYOK — bring your own key. You can plug in any LLM provider. But pair it with Venice and you get the full stack: a censorship-resistant AI model running inference with no content filtering, connected to an agent whose state lives on a decentralized network that no company can shut down, with end-to-end encrypted conversations that nobody can read — not the relay operator, not the LLM provider, not the blockchain validators. Venice doesn't log prompts. Flow can't read your encrypted storage. The relay never sees your plaintext. That's not a privacy policy. That's architecture. You can also use OpenAI, Anthropic, or any OpenAI-compatible provider. The agent platform doesn't care — it's model-agnostic. But the Venice pairing is the one that closes every gap in the stack. For the people tinkering with OpenClaw and the broader open-source agent ecosystem — FlowClaw is exploring what happens when you take the agent off the cloud entirely. Not just open-sourcing the code (though it is), but putting the actual runtime state on a distributed computer. Your agent's memory isn't in a SQLite file on your laptop or a Pinecone index on someone's cluster. It's on-chain, encrypted, and replicated across every validator node on Flow. You own it the way you own a private key — mathematically, not contractually. The blockchain here isn't a gimmick bolted onto an agent for token speculation. It's functioning as the infrastructure layer that replaces AWS. Flow accounts are programmable containers with their own storage, keys, and security capabilities. Passkey authentication works natively because Flow supports P-256 keys at the protocol level — the same curve your phone uses for biometrics. Gas sponsorship works natively because Flow transactions have separate proposer, authorizer, and payer roles built into the protocol. No proxy contracts. No relayers. No ERC-4337. Now here's the part that interests me economically. Every FlowClaw interaction is an on-chain transaction. Every message stored, every memory committed, every session created, every sub-agent spawned. An active user might generate dozens of transactions in a single conversation. Scale that and FlowClaw becomes a real contributor to Flow's transaction volume. Flow.com becomes deflationary at 250 TPS. Applications like FlowClaw that generate high-frequency, storage-heavy transactions are exactly what moves the needle. Every encrypted message uses account storage, which requires FLOW balance to back it. Every transaction burns fees. The more agents running, the more demand for $FLOW — not because of a tokenomics gimmick, but because the protocol literally requires it for compute and storage. FlowClaw doesn't have its own token. The token is $FLOW. The entire platform runs natively on the network — using Flow storage, paying Flow transaction fees, backed by Flow account balances. If FlowClaw succeeds, FLOW captures that value directly. I'm sharing this early because the AI agent space is moving fast and I think the decentralized infrastructure angle is underexplored. Most "crypto AI" projects are tokens with a chatbot attached. FlowClaw is the opposite — it's an agent platform that happens to use a blockchain because the blockchain solves real engineering problems that centralized infrastructure can't. Try it: Github: Create an agent, ask it something, spawn a sub-agent, check your memory tab, pair it with Venice for the full censorship-resistant stack. Break it and tell me what broke. If you think this direction matters, the best thing you can do is use it and give feedback. Your AI agent should be yours. Not your provider's. Not your platform's. Yours.

doodlifts

12,166 views • 5 months ago

Well, here goes. Our biggest announcement ever: Social Snowball ☃️ now integrates with TikTok Shop! 🎉 The future of affiliate marketing is creators, but nothing is worse than trying to manage thousands of creator partnerships distributed between multiple platforms and social channels. This native integration will change the way you manage your creator partnerships, forever: ⏰ Save time with unified affiliate management: View an affiliate's TikTok Shop and DTC affiliate data from one unified profile. 📣 Recruit more TikTok Shop affiliates on autopilot: Enroll your DTC affiliates into a collaboration on TikTok Shop right from their affiliate dashboard. 📲 Reduce friction with unified enrollment: Create one sign up flow to enroll affiliates in both your DTC and TikTok Shop affiliate programs. 🚀 Boost GMV with gamification & tiers: Let affiliates unlock gifts and higher commissions for achieving milestones and hitting GMV targets. Some of the top TikTok Shop brands like Javy Coffee and Smackin Snacks are already using this, and now you can too! This is live now for all existing and new brands on Social Snowball (Blizzard and Enterprise plans only). This integration has been a long time coming, and it has been an absolute pleasure collaborating with the TikTok Shop team on this. Also need to give a massive shoutout to the entire Social Snowball team for bringing this to life on a tight deadline. This is just the beginning, and we will be rolling out even more functionality for TikTok Shop over the next few months so stay tuned! Learn more here:

Noah Tucker ☃️

88,101 views • 1 year ago