Video yükleniyor...

Video Yüklenemedi

Bu video yüklenirken bir sorun oluştu. Bu geçici bir ağ sorunundan kaynaklanıyor olabilir veya video kullanılamıyor olabilir.

Ana Sayfaya Dön

Human3D 🧑‍🤝‍🧑 "3D Segmentation of Humans in Point Clouds with Synthetic Data" was accepted at #ICCV2025 🥳 Paper: Project page: #ICCV2023 Ayça Takmaz Jonas Schult Siyu Tang @VLG-ETHZ

Francis Engelmann

1,651 subscribers

13,829 görüntüleme • 3 yıl önce •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Yorum

Yorum bulunmuyor

Orijinal gönderinin yorumları burada görünecek

Benzer Videolar

Coarse2Real (C2R) transfers simple 3D renderings into realistic style video. Check our paper and project page to learn how to hedge small amount of synthetic paired data with real non-pair data for training the C2R model. We will release the model soon!

Coarse2Real (C2R) transfers simple 3D renderings into realistic style video. Check our paper and project page to learn how to hedge small amount of synthetic paired data with real non-pair data for training the C2R model. We will release the model soon!

Yi Zhou

12,406 görüntüleme • 4 ay önce

Trained on zero real-world data. Learned to walk, pick up boxes, and follow multi-step instructions... in the REAL world. ( 📌 Paper below) Researchers from Amazon FAR, Berkeley, Stanford, and CMU scanned real rooms with an iPhone, rebuilt them as 3D Gaussian Splatting scenes, then generated 48,000 synthetic trajectories of a Unitree G1 walking, grasping, and placing objects inside those virtual replicas. They rendered the robot's first-person camera view from each run and paired it with the matching language instruction and motion data. That's the dataset every humanoid team needs and nobody has: synced egocentric video + language + kinematics, at scale. Instead of collecting it in the real world, they manufactured it. They trained a vision-language-kinematics policy on that synthetic data alone, then deployed it on the physical G1 across five task types: navigation to a named object, lifting boxes of three different sizes with no per-size tuning, chained multi-step tasks, robustness to mid-task layout changes and flickering lights, and multi-minute long-horizon runs. No real-world fine-tuning at any point. Real-world interaction data has been the hard limit on humanoid learning... slow, expensive, and small. If scanning a room once and synthesizing thousands of labeled interactions holds up as a general recipe, that limit moves. Data stops being the bottleneck robotics teams have to solve for. 📌 Paper: Project: ——- Weekly robotics and AI insights. Subscribe free:

Trained on zero real-world data. Learned to walk, pick up boxes, and follow multi-step instructions... in the REAL world. ( 📌 Paper below) Researchers from Amazon FAR, Berkeley, Stanford, and CMU scanned real rooms with an iPhone, rebuilt them as 3D Gaussian Splatting scenes, then generated 48,000 synthetic trajectories of a Unitree G1 walking, grasping, and placing objects inside those virtual replicas. They rendered the robot's first-person camera view from each run and paired it with the matching language instruction and motion data. That's the dataset every humanoid team needs and nobody has: synced egocentric video + language + kinematics, at scale. Instead of collecting it in the real world, they manufactured it. They trained a vision-language-kinematics policy on that synthetic data alone, then deployed it on the physical G1 across five task types: navigation to a named object, lifting boxes of three different sizes with no per-size tuning, chained multi-step tasks, robustness to mid-task layout changes and flickering lights, and multi-minute long-horizon runs. No real-world fine-tuning at any point. Real-world interaction data has been the hard limit on humanoid learning... slow, expensive, and small. If scanning a room once and synthesizing thousands of labeled interactions holds up as a general recipe, that limit moves. Data stops being the bottleneck robotics teams have to solve for. 📌 Paper: Project: ——- Weekly robotics and AI insights. Subscribe free:

Ilir Aliu

12,950 görüntüleme • 1 ay önce

Human3R: Everyone Everywhere All at Once Note: I recorded the video from the interactive demo on their project page (linked in the comment below). Abstract (excerpt): Human3R jointly recovers global multi-person SMPL-X bodies ("everyone"), dense 3D scenes ("everywhere"), and camera trajectories in a single forward pass ("all-at-once"). Our method builds upon the 4D online reconstruction model CUT3R and uses parameter-efficient visual prompt tuning to preserve CUT3R's rich spatiotemporal priors while enabling direct readout of multiple SMPL-X bodies. Human3R is a unified model that eliminates heavy dependencies and iterative refinement. After being trained on the relatively small-scale synthetic dataset BEDLAM for just one day on one GPU, it achieves superior performance with remarkable efficiency: it reconstructs multiple humans in a one-shot manner, along with 3D scenes, in one stage, at real-time speed (15 FPS) with a low memory footprint (8 GB).

Human3R: Everyone Everywhere All at Once Note: I recorded the video from the interactive demo on their project page (linked in the comment below). Abstract (excerpt): Human3R jointly recovers global multi-person SMPL-X bodies ("everyone"), dense 3D scenes ("everywhere"), and camera trajectories in a single forward pass ("all-at-once"). Our method builds upon the 4D online reconstruction model CUT3R and uses parameter-efficient visual prompt tuning to preserve CUT3R's rich spatiotemporal priors while enabling direct readout of multiple SMPL-X bodies. Human3R is a unified model that eliminates heavy dependencies and iterative refinement. After being trained on the relatively small-scale synthetic dataset BEDLAM for just one day on one GPU, it achieves superior performance with remarkable efficiency: it reconstructs multiple humans in a one-shot manner, along with 3D scenes, in one stage, at real-time speed (15 FPS) with a low memory footprint (8 GB).

MrNeRF

35,861 görüntüleme • 11 ay önce

Check out this Stereo4D paper from Google DeepMind. It's a pretty clever approach to a persistent problem in computer vision -- getting good training data for how things move in 3D. The key insight is using VR180 videos -- those stereo fisheye videos we launched back in 2017 for YouTubeVR. It was always clear that structured stereo datasets would be valuable for computer vision -- and we launched some powerful VR tools with it back in 2017 (link below). But what's the game changer now in 2024 is the scale -- they're providing 110K high quality clips :-) That's the kind of massive, real-world AI dataset that was just a dream back then! They're using it to train this model called DynaDUSt3R that can predict both 3D structure and motion from video frames. Which means it tracks how objects move between frames while simultaneously reconstructing their 3D shape. And given we're dealing with real stereoscopic content, results are notably better than synthetic data, giving you a faithful rendition of the real-world with a diverse set of subject matter. It's one of those through lines when tackling a timeless mission like mapping the world or spatial computing -- VR content created for immersion becoming the foundation for teaching machines to understand how the world moves. Sometimes innovation chains together in unexpected ways! Links to projects below⛓️

Check out this Stereo4D paper from Google DeepMind. It's a pretty clever approach to a persistent problem in computer vision -- getting good training data for how things move in 3D. The key insight is using VR180 videos -- those stereo fisheye videos we launched back in 2017 for YouTubeVR. It was always clear that structured stereo datasets would be valuable for computer vision -- and we launched some powerful VR tools with it back in 2017 (link below). But what's the game changer now in 2024 is the scale -- they're providing 110K high quality clips :-) That's the kind of massive, real-world AI dataset that was just a dream back then! They're using it to train this model called DynaDUSt3R that can predict both 3D structure and motion from video frames. Which means it tracks how objects move between frames while simultaneously reconstructing their 3D shape. And given we're dealing with real stereoscopic content, results are notably better than synthetic data, giving you a faithful rendition of the real-world with a diverse set of subject matter. It's one of those through lines when tackling a timeless mission like mapping the world or spatial computing -- VR content created for immersion becoming the foundation for teaching machines to understand how the world moves. Sometimes innovation chains together in unexpected ways! Links to projects below⛓️

Bilawal Sidhu

68,515 görüntüleme • 1 yıl önce

This isn’t a 3D model file. It’s 100% code Our world-famous Mars Cat avatar has been reconstructed entirely as Three.js source code No GLB at runtime. No model download. No textures or UV atlas. Just code, rendered directly in the browser It was created with IMG2THREEJS by Nick. An incredible open-source project with over 15,000 stars on GitHub And our Mars Cat is now featured on the home page gallery See it live below ↓. Grab the code. Add it to your game New to Three.js? It’s one of the most popular open-source JavaScript libraries for building interactive 3D experiences and games on the web

Jay B 😻🚀

15,044 görüntüleme • 11 gün önce

You can't 3D reconstruct glass from images... ...WRONG! Thanks for video diffusion, now just about anything is possible! Introducing...Diffusion Knows Transparency (DKT) Transparent and reflective objects usually break robot vision and photogrammetry pipelines because they don't follow the "solid object" rules standard cameras expect. DKT is a new AI model that repurposes the "internal physics engine" found in video generation models to solve this problem. Researchers took a massive video diffusion model (WAN) and fine-tuned it using a custom-built synthetic dataset to turn it into a high-precision depth sensor. To train the AI, they built the first massive synthetic video library of transparent objects, 1.32 million frames of perfectly labeled glass and metal objects in motion. Without ever seeing a "real" labeled video of glass during training, the model (DKT) outperformed all previous specialized systems on real-world benchmarks (ClearPose, DREDS). They created a "lightweight" 1.3B parameter version that runs fast enough (0.17s per frame) to be used on actual robot hardware. Two reasons I find this project important: 1. It further proves that synthetic data will be essential for training the next generation vision models. 2. In real-world robotic tests, using DKT's depth maps nearly doubled the success rate of robot arms trying to pick up objects on tricky reflective or translucent surfaces. At home robots will need to interact with these types of objects on a daily basis. Check out the project page here: Code is LIVE! #Computervision #Robotics #AI

You can't 3D reconstruct glass from images... ...WRONG! Thanks for video diffusion, now just about anything is possible! Introducing...Diffusion Knows Transparency (DKT) Transparent and reflective objects usually break robot vision and photogrammetry pipelines because they don't follow the "solid object" rules standard cameras expect. DKT is a new AI model that repurposes the "internal physics engine" found in video generation models to solve this problem. Researchers took a massive video diffusion model (WAN) and fine-tuned it using a custom-built synthetic dataset to turn it into a high-precision depth sensor. To train the AI, they built the first massive synthetic video library of transparent objects, 1.32 million frames of perfectly labeled glass and metal objects in motion. Without ever seeing a "real" labeled video of glass during training, the model (DKT) outperformed all previous specialized systems on real-world benchmarks (ClearPose, DREDS). They created a "lightweight" 1.3B parameter version that runs fast enough (0.17s per frame) to be used on actual robot hardware. Two reasons I find this project important: 1. It further proves that synthetic data will be essential for training the next generation vision models. 2. In real-world robotic tests, using DKT's depth maps nearly doubled the success rate of robot arms trying to pick up objects on tricky reflective or translucent surfaces. At home robots will need to interact with these types of objects on a daily basis. Check out the project page here: Code is LIVE! #Computervision #Robotics #AI

Jonathan Stephens

17,712 görüntüleme • 8 ay önce

New Cell paper from Bergles lab at Johns Hopkins just built the most comprehensive map of brain myelin ever made — every oligodendrocyte, across the entire mouse brain, across the lifespan. The scale: >10 million cells per brain, terabyte-scale 3D lightsheet volumes, registered to the Allen Brain Atlas across 417 regions from 2 months to 2+ years of age. The technical stack: Custom tissue clearing (CUBIC-L + SHIELD + uRIMS with 40% urea) to preserve endogenous fluorescence. 3D Mask R-CNN for instance segmentation — not just semantic, instance — so it can distinguish individual cells within dense clusters at scale via overlapping sliding windows. Vision Transformer to classify newly-formed vs. mature oligodendrocytes using soma morphology. All cross-referenced against Allen ISH transcriptomics and MICrONS serial EM. What they found: Oligodendrocyte density varies 10,000-fold across brain regions. Left-right hemispheres: r=0.99. Sex: no significant difference. Strain: matters. The brain never stops myelinating. New oligodendrocytes are still being generated in 2-year-old mice. Prefrontal cortex L6 shows the fastest rates of new myelination into old age — the circuits for executive function keep rewiring throughout life. After demyelination, L4 sensory cortex is the most resilient — oligodendrocytes survive at higher rates. The hippocampus loses nearly everything and barely recovers. Degree of injury doesn't predict rate of recovery. These are independent axes. The Alzheimer's result is the most surprising: Dense-core plaques dominate in cortex and hippocampus. Diffuse/small-core plaques dominate in white matter fiber tracts. Old assumption: diffuse plaques are "less toxic." The data says the opposite — small plaques in fiber tracts cause more myelin loss per plaque than dense-core plaques in gray matter. Plaque load and oligodendrocyte loss are essentially uncorrelated (ρ=0.22). The damage is plaque-type and location specific, not load-dependent. For MS and AD research: you can't read off white matter injury from gray matter plaque burden. The pathology in fiber tracts is running on different rules. Data: Paper:

New Cell paper from Bergles lab at Johns Hopkins just built the most comprehensive map of brain myelin ever made — every oligodendrocyte, across the entire mouse brain, across the lifespan. The scale: >10 million cells per brain, terabyte-scale 3D lightsheet volumes, registered to the Allen Brain Atlas across 417 regions from 2 months to 2+ years of age. The technical stack: Custom tissue clearing (CUBIC-L + SHIELD + uRIMS with 40% urea) to preserve endogenous fluorescence. 3D Mask R-CNN for instance segmentation — not just semantic, instance — so it can distinguish individual cells within dense clusters at scale via overlapping sliding windows. Vision Transformer to classify newly-formed vs. mature oligodendrocytes using soma morphology. All cross-referenced against Allen ISH transcriptomics and MICrONS serial EM. What they found: Oligodendrocyte density varies 10,000-fold across brain regions. Left-right hemispheres: r=0.99. Sex: no significant difference. Strain: matters. The brain never stops myelinating. New oligodendrocytes are still being generated in 2-year-old mice. Prefrontal cortex L6 shows the fastest rates of new myelination into old age — the circuits for executive function keep rewiring throughout life. After demyelination, L4 sensory cortex is the most resilient — oligodendrocytes survive at higher rates. The hippocampus loses nearly everything and barely recovers. Degree of injury doesn't predict rate of recovery. These are independent axes. The Alzheimer's result is the most surprising: Dense-core plaques dominate in cortex and hippocampus. Diffuse/small-core plaques dominate in white matter fiber tracts. Old assumption: diffuse plaques are "less toxic." The data says the opposite — small plaques in fiber tracts cause more myelin loss per plaque than dense-core plaques in gray matter. Plaque load and oligodendrocyte loss are essentially uncorrelated (ρ=0.22). The damage is plaque-type and location specific, not load-dependent. For MS and AD research: you can't read off white matter injury from gray matter plaque burden. The pathology in fiber tracts is running on different rules. Data: Paper:

Bo Wang

24,807 görüntüleme • 6 ay önce

𝗗𝗼𝗻'𝘁 𝗳𝗶𝗻𝗲-𝘁𝘂𝗻𝗲 𝗿𝗼𝗯𝗼𝘁 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹𝘀. 𝗦𝘁𝗲𝗲𝗿 𝘁𝗵𝗲𝗺 𝘄𝗶𝘁𝗵 𝗵𝘂𝗺𝗮𝗻 𝗰𝗼𝗿𝗿𝗲𝗰𝘁𝗶𝗼𝗻𝘀 𝗶𝗻𝘀𝘁𝗲𝗮𝗱, 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗰𝗵𝗮𝗻𝗴𝗶𝗻𝗴 𝘁𝗵𝗲 𝗯𝗮𝘀𝗲 𝗽𝗼𝗹𝗶𝗰𝘆 Modern VLAs and world-action models can perform impressive manipulation skills, but adapting them reliably to new robots and tasks remains challenging. A natural solution is DAgger-style online imitation learning: deploy the robot, collect human corrections, and update the policy. Yet foundation models are fragile in the low-data regime, fine-tuning on a handful of interventions can improve one behavior while degrading others. Online post-training or reinforcement learning can require costly data collection and exploration, making real-world learning expensive and potentially unsafe. In our new paper, 𝗙𝗹𝗼𝘄𝗗𝗔𝗴𝗴𝗲𝗿, we take a different approach: 𝗜𝗻𝘀𝘁𝗲𝗮𝗱 𝗼𝗳 𝗰𝗵𝗮𝗻𝗴𝗶𝗻𝗴 𝘁𝗵𝗲 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹, 𝘄𝗲 𝗹𝗲𝗮𝗿𝗻 𝗵𝗼𝘄 𝘁𝗼 𝘀𝘁𝗲𝗲𝗿 𝗶𝘁 𝗳𝗿𝗼𝗺 𝗵𝘂𝗺𝗮𝗻 𝗰𝗼𝗿𝗿𝗲𝗰𝘁𝗶𝗼𝗻𝘀. The key idea is 𝗮𝗰𝘁𝗶𝗼𝗻 𝗶𝗻𝘃𝗲𝗿𝘀𝗶𝗼𝗻: we map human corrective actions back into the latent noise space of the frozen generative policy. These latent targets train a lightweight controller that adapts the robot while preserving the original model's capabilities. Across simulation and real robots, FlowDAgger: 📈 Learns from only 5–20 human intervention episodes 🏆 Outperforms supervised fine-tuning and latent-space reinforcement learning 🤖 Works across VLAs, diffusion policies, and world-action models ✔️ Provides reliable improvements without modifying the pretrained policy We believe this offers a practical path toward making robot foundation models improve during deployment, learning from the way humans naturally teach: through corrections. 📄 Paper: 🌐 Project: 💻 Code: This project was led by my amazing colleague Michael Murray with help from Daphne Chen, Simran Bagaria, Dean Fortier, Tess Hellebrekers, Harshavardhan Reddy Gajarla, Galen Mullins and Andrey Kolobov at Microsoft Research and Maya Cakmak at University of Washington

𝗗𝗼𝗻'𝘁 𝗳𝗶𝗻𝗲-𝘁𝘂𝗻𝗲 𝗿𝗼𝗯𝗼𝘁 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹𝘀. 𝗦𝘁𝗲𝗲𝗿 𝘁𝗵𝗲𝗺 𝘄𝗶𝘁𝗵 𝗵𝘂𝗺𝗮𝗻 𝗰𝗼𝗿𝗿𝗲𝗰𝘁𝗶𝗼𝗻𝘀 𝗶𝗻𝘀𝘁𝗲𝗮𝗱, 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗰𝗵𝗮𝗻𝗴𝗶𝗻𝗴 𝘁𝗵𝗲 𝗯𝗮𝘀𝗲 𝗽𝗼𝗹𝗶𝗰𝘆 Modern VLAs and world-action models can perform impressive manipulation skills, but adapting them reliably to new robots and tasks remains challenging. A natural solution is DAgger-style online imitation learning: deploy the robot, collect human corrections, and update the policy. Yet foundation models are fragile in the low-data regime, fine-tuning on a handful of interventions can improve one behavior while degrading others. Online post-training or reinforcement learning can require costly data collection and exploration, making real-world learning expensive and potentially unsafe. In our new paper, 𝗙𝗹𝗼𝘄𝗗𝗔𝗴𝗴𝗲𝗿, we take a different approach: 𝗜𝗻𝘀𝘁𝗲𝗮𝗱 𝗼𝗳 𝗰𝗵𝗮𝗻𝗴𝗶𝗻𝗴 𝘁𝗵𝗲 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹, 𝘄𝗲 𝗹𝗲𝗮𝗿𝗻 𝗵𝗼𝘄 𝘁𝗼 𝘀𝘁𝗲𝗲𝗿 𝗶𝘁 𝗳𝗿𝗼𝗺 𝗵𝘂𝗺𝗮𝗻 𝗰𝗼𝗿𝗿𝗲𝗰𝘁𝗶𝗼𝗻𝘀. The key idea is 𝗮𝗰𝘁𝗶𝗼𝗻 𝗶𝗻𝘃𝗲𝗿𝘀𝗶𝗼𝗻: we map human corrective actions back into the latent noise space of the frozen generative policy. These latent targets train a lightweight controller that adapts the robot while preserving the original model's capabilities. Across simulation and real robots, FlowDAgger: 📈 Learns from only 5–20 human intervention episodes 🏆 Outperforms supervised fine-tuning and latent-space reinforcement learning 🤖 Works across VLAs, diffusion policies, and world-action models ✔️ Provides reliable improvements without modifying the pretrained policy We believe this offers a practical path toward making robot foundation models improve during deployment, learning from the way humans naturally teach: through corrections. 📄 Paper: 🌐 Project: 💻 Code: This project was led by my amazing colleague Michael Murray with help from Daphne Chen, Simran Bagaria, Dean Fortier, Tess Hellebrekers, Harshavardhan Reddy Gajarla, Galen Mullins and Andrey Kolobov at Microsoft Research and Maya Cakmak at University of Washington

Oier Mees

13,276 görüntüleme • 2 ay önce

save these two free services. they are what i built my Polymarket strategies on, the ones paying me around $10,000 a month. what it looks like day to day: you ask any question about the market in plain words, Claude Code digs through the data and comes back with numbers. you become a quant researcher. 1 - poly_data one command and every Polymarket trade sits on your own disk. who bought, when, at what price, what size, and who was standing on the other side of it. my first sync pulled 3.1 million markets. you can pull apart the edge of any trader on Polymarket you like. first sync runs a few hours, leave it overnight, after that updates land in seconds. 2 - pendulumflow archive this one holds the book itself: what orders were sitting there, at what prices, what size, second by second, on every single market. 288 billion events recorded, and 2.3 billion more coming in every day. i used to pay $200 a month for this kind of data. here you just open it and take it, nothing to sign up for. the front page has ready made questions you copy straight into Claude Code. the project runs on donations, storing and serving this much data costs real money every month. so if you start using it, send the guy something, he is carrying all of it alone. these two will give you what only a handful of people on this market have.

save these two free services. they are what i built my Polymarket strategies on, the ones paying me around $10,000 a month. what it looks like day to day: you ask any question about the market in plain words, Claude Code digs through the data and comes back with numbers. you become a quant researcher. 1 - poly_data one command and every Polymarket trade sits on your own disk. who bought, when, at what price, what size, and who was standing on the other side of it. my first sync pulled 3.1 million markets. you can pull apart the edge of any trader on Polymarket you like. first sync runs a few hours, leave it overnight, after that updates land in seconds. 2 - pendulumflow archive this one holds the book itself: what orders were sitting there, at what prices, what size, second by second, on every single market. 288 billion events recorded, and 2.3 billion more coming in every day. i used to pay $200 a month for this kind of data. here you just open it and take it, nothing to sign up for. the front page has ready made questions you copy straight into Claude Code. the project runs on donations, storing and serving this much data costs real money every month. so if you start using it, send the guy something, he is carrying all of it alone. these two will give you what only a handful of people on this market have.

may.crypto {🦅}

25,181 görüntüleme • 6 gün önce

LinkedIn wrapped is here! And it's made with Rive Stoked to be invited by BUCK to join the project. I wasn’t actually animating the thing though I had more of a technical assignment and we had quite a few things to solve. There is 9 chapters total (but not everyone sees all the chapters or even all the slides in given chapter) in 3 languages, the data being sent to Rive is raw so we had to cover for many variables: - A progress bar aware what chapters are included for given user - A progress bar checking what slides are included for and logging their duration time to progress bar - A custom slider solution (the default Rive slider is not enough if you turn contents on and off) - A State Machine logic aware of chapters or slides skipped from playback - A custom slider behavior for when there is auto-slide (instant) and tap-slide (slide animation) - A custom text-length tracker for to dynamically change font-size for long strings - A custom screen ratio tracker to scale down the UI for small screens - Handling dynamically handle big numbers (10,000 becomes 10k etc.) - Handling translations (3 languages) - Super complex behaviour for when if some data-point is missing we don’t leave a blank space, but it is being replaced with other data point instead (so some user se A-B-C, but some will see only B-C as if A never existed without leaving blank space) - Rage click prevention Amazing experience, thanks for having me. People at BUCK are absolutely goated, kudos to everyone who was helping me and to everyone who was actually doing the animations, stunning. Is it the biggest Rive exposure yet? Maybe Rive

LinkedIn wrapped is here! And it's made with Rive Stoked to be invited by BUCK to join the project. I wasn’t actually animating the thing though I had more of a technical assignment and we had quite a few things to solve. There is 9 chapters total (but not everyone sees all the chapters or even all the slides in given chapter) in 3 languages, the data being sent to Rive is raw so we had to cover for many variables: - A progress bar aware what chapters are included for given user - A progress bar checking what slides are included for and logging their duration time to progress bar - A custom slider solution (the default Rive slider is not enough if you turn contents on and off) - A State Machine logic aware of chapters or slides skipped from playback - A custom slider behavior for when there is auto-slide (instant) and tap-slide (slide animation) - A custom text-length tracker for to dynamically change font-size for long strings - A custom screen ratio tracker to scale down the UI for small screens - Handling dynamically handle big numbers (10,000 becomes 10k etc.) - Handling translations (3 languages) - Super complex behaviour for when if some data-point is missing we don’t leave a blank space, but it is being replaced with other data point instead (so some user se A-B-C, but some will see only B-C as if A never existed without leaving blank space) - Rage click prevention Amazing experience, thanks for having me. People at BUCK are absolutely goated, kudos to everyone who was helping me and to everyone who was actually doing the animations, stunning. Is it the biggest Rive exposure yet? Maybe Rive

Bartek Radziejewski

24,978 görüntüleme • 9 ay önce

I went a little overboard with Codex last week and burned through my entire weekly allowance in two days. Luckily, my quota reset today. Otherwise, I’m not sure what I would’ve done. It got me thinking: instead of asking one large model to handle everything from start to finish, why not let a stronger model plan the project and review the work, while a model built for execution handles the day-to-day implementation? So I tried it. The result was better than I expected. I used GPT-5.6 Sol in Codex as the decision-maker, then ran Ling-3.0-flash from Ant Ling inside OpenCode as the execution engine. Together, they built a small 3D farming game. Before writing any code, I had Codex create four documents: SPEC.md defined the product scope and the lines we couldn’t cross. ARCHITECTURE.md laid out the isometric coordinate system, state machine, and module boundaries. TASKS.md broke the project into small jobs Ling could tackle one at a time. ACCEPTANCE.md explained how each step would be tested and what “done” actually meant. Then I gave Ling a very straightforward role: You are the execution model for this project. Read all four documents before you begin. Work only on the task assigned for this round. When you’re done, run typecheck, test, and build. If anything fails, read the error, fix it, and run the checks again. Do not move on to the next task early. Ling handled dependency installation, project structure, strict TypeScript configuration, test setup, and a production build in 6 minutes and 3 seconds. It ran into issues with the Vite test config, a TS6310 error, and a missing jsdom dependency along the way. Instead of stopping at the first error, it kept reading the logs and fixing the problems until all three checks passed. The speed was honestly hard to believe. If you exclude the time spent waiting on tools, it was producing more than 100 tokens per second. That made the whole development loop feel noticeably faster. After this experiment, I’m planning to keep using the same workflow. If the task is small, there’s no reason to call an expensive planning model for every single step. If the task is large, handing the entire project to a Flash model in one prompt isn’t a great idea either. The setup that makes more sense to me is: Use a more capable model such as Codex to explore the project, make architectural decisions, and break the work down. Put the constraints into specs, schemas, types, and tests instead of leaving them buried in chat history. Give Ling-3.0-flash a steady stream of clear, verifiable implementation tasks. Report bugs with structured context and actual error logs, rather than saying, “It still doesn’t work.” Bring Codex back in for architecture reviews, visual checks, and changes that affect multiple parts of the project. The point of this setup isn’t to give AI a big “build the whole project” button. It’s to turn software development into a pipeline with a much more sensible cost structure: Codex figures out the plan, sets the boundaries, and catches problems. Ling-3.0-flash moves quickly, calls tools reliably, and works through well-defined tasks at scale. For agent workflows that involve lots of repetitive edits, production tasks, and tool calls, this may be a more practical answer than simply using the biggest model for everything.

I went a little overboard with Codex last week and burned through my entire weekly allowance in two days. Luckily, my quota reset today. Otherwise, I’m not sure what I would’ve done. It got me thinking: instead of asking one large model to handle everything from start to finish, why not let a stronger model plan the project and review the work, while a model built for execution handles the day-to-day implementation? So I tried it. The result was better than I expected. I used GPT-5.6 Sol in Codex as the decision-maker, then ran Ling-3.0-flash from Ant Ling inside OpenCode as the execution engine. Together, they built a small 3D farming game. Before writing any code, I had Codex create four documents: SPEC.md defined the product scope and the lines we couldn’t cross. ARCHITECTURE.md laid out the isometric coordinate system, state machine, and module boundaries. TASKS.md broke the project into small jobs Ling could tackle one at a time. ACCEPTANCE.md explained how each step would be tested and what “done” actually meant. Then I gave Ling a very straightforward role: You are the execution model for this project. Read all four documents before you begin. Work only on the task assigned for this round. When you’re done, run typecheck, test, and build. If anything fails, read the error, fix it, and run the checks again. Do not move on to the next task early. Ling handled dependency installation, project structure, strict TypeScript configuration, test setup, and a production build in 6 minutes and 3 seconds. It ran into issues with the Vite test config, a TS6310 error, and a missing jsdom dependency along the way. Instead of stopping at the first error, it kept reading the logs and fixing the problems until all three checks passed. The speed was honestly hard to believe. If you exclude the time spent waiting on tools, it was producing more than 100 tokens per second. That made the whole development loop feel noticeably faster. After this experiment, I’m planning to keep using the same workflow. If the task is small, there’s no reason to call an expensive planning model for every single step. If the task is large, handing the entire project to a Flash model in one prompt isn’t a great idea either. The setup that makes more sense to me is: Use a more capable model such as Codex to explore the project, make architectural decisions, and break the work down. Put the constraints into specs, schemas, types, and tests instead of leaving them buried in chat history. Give Ling-3.0-flash a steady stream of clear, verifiable implementation tasks. Report bugs with structured context and actual error logs, rather than saying, “It still doesn’t work.” Bring Codex back in for architecture reviews, visual checks, and changes that affect multiple parts of the project. The point of this setup isn’t to give AI a big “build the whole project” button. It’s to turn software development into a pipeline with a much more sensible cost structure: Codex figures out the plan, sets the boundaries, and catches problems. Ling-3.0-flash moves quickly, calls tools reliably, and works through well-defined tasks at scale. For agent workflows that involve lots of repetitive edits, production tasks, and tool calls, this may be a more practical answer than simply using the biggest model for everything.

雪踏乌云

23,107 görüntüleme • 1 ay önce

A case of two coronary scans: Nick Norwitz (high-fat, keto 🍳); Simon Hill (plant-based 🌱) First, for context, this post is largely in response to an incredibly nuanced and non-dogmatic podcast conversation between Dave Feldman and Darius Sharpe, one that had me scratching my chin and holding my ribs throughout. I definitely recommend giving it a listen. 12/10 Episode! They spend a substantial amount of time discussing cardiovascular risk and, in particular, the contrast between my coronary CT scan, showing 0 calcified and 0 soft plaque (total plaque 0mm3) despite astronomical LDL levels in the high 500s for seven years vs that of plant-based nutrition influencer Simon Hill, which showed 61.3 mm3 of plaque despite his never having had high LDL and, for the last decade or so, targeting exceptionally low LDL and ApoB, including by taking a PCSK9 inhibitor and eating a low-saturated-fat, plant-forward diet. First and foremost, there’s only so much you can conclude from a comparison of two cases. That’s obvious, so let me state that clearly. However, considering the broader context and expressed public interested, it’s still an interesting thought puzzle: What is “protecting” me from atherosclerosis, or what is leading Simon to develop plaque despite his low LDL and ApoB? I’ll address the common hand-wavy answer up front: “Well, maybe Nick has some special protective genetics.” What I’d say to this is that, although I can’t rule it out definitively since we haven’t annotated the entire human genome, my father had 99% occlusion of his left anterior descending artery at 44, and I also inherited his high Lp(a), which sits between 150 and 200. That is in addition to my historically exceptionally high LDL and a history of inflammatory bowel disease, which should further increase my risk of progression. Yet despite all of that, I have no measurable plaque, even though, as the patient, I can tell you that many cardiologists assured me I would. Additionally, the excuse that “he’s just young” doesn’t fly, since Simon is also young. We’re both in our 30s. And, as a historical comparator, children with homozygous familial hypercholesterolemia can develop plaque within a few years of being born. So, puzzle me this: Why am I not developing plaque if the relationship between LDL/ApoB exposure and plaque progression is so "remarkably consistent?" Hand-wave if you want, but I’d like something more specific please. (And I’d like anybody to tell me why they think I was confident enough to spend seven years on a self-experiment, risking my heart on the bet that this would effectively be the outcome. Was I just cocky and lucky?) That’s mostly what I want to post: a thought question to consider based on two individuals who have chosen to make their data publicly available. However, since this field is so wrapped in controversy, let me elaborate on a few more points that are related. 1. KETO-CTA. To anyone who wants to pivot and try to point the finger at KETO-CTA, I will remind you that, in 4 separate analyses, there was no relationship between LDL or ApoB and plaque progression. That remains true to this day. In fact, it's a result that remains more robust for having had our hands forced to run the additional analyses. With respect to our retraction of the April 7th Cleerly paper, which continues to be misrepresented (at this point, I have to assume willfully) I will remind you: i) we the authors, pushed forward the retraction of the April 7 paper; it was not forced upon us ii) we did so because of the Cleerly anomalies that came to light after publication – nb: Dave and I were blinded to elements of data willfully to protect the integrity of the project (there’s an irony for you) iii) we sought retraction only after company’s refusal to repeat a properly blinded rescan and quality assessment, and subsequent analyses that made the only reasonable interpretation that Cleerly systematically overestimated plaque progression iv) in none of the analyses did ApoB or LDL predict plaque progression, which was the main novel finding and remains robust. 2. Ezetimibe I am now taking ezetimibe. Critics have tried to cast this - incorrectly - as some form of me hedging, but I intentionally did not start it until after my 7-year scan, and I am not withholding any information from or about that scan. The reason I’m taking ezetimibe, which I have very clearly disclosed, including in my newsletter and in a video, is 2024 Aging Biology research suggesting cholesterol-independent effects on neuroprotection, which is something that’s important to me. The fact that I am taking it, I would cast as a testament that I’m not dogmatic on this issue, and I’m not trying to “LDL-maxx” for some aware. Rather, I’m being open, honest and transparent in my journey as a person and patient trying to make the best decision for myself at ever step of the journey. Also, on the topic of ezetimibe, see recent work by Adrian Soto-Mota DoctorTro et al., which adds another layer of nuance to LDL and ApoB management in the context of low-carbohydrate diets. TL;DR, if someone sees their LDL and ApoB rise on a low-carbohydrate / ketogenic diet, ezetimibe is likely to punch above its weight for LDL and ApoB reduction. Setting aside dogma, understanding individual biology and how to manipulate it can often be very useful in a clinical setting. I will always put data first. Acknowledging this fact is not contradictory to anything I’ve said, but completely consistent with it. It is all about individual context. Now, everyone should go listen to episode 43 of TFP_ with Dave Feldman and Darius Sharpe (Spotify, Apple and YouTube). It’s stupendous! P.S. I want to thank Dave for sharing on the podcast that, before each of my cardiac scans, I said a priori that I would release the results, whatever they were. You all know Dave to be impeccably honest. That is not in dispute, and it has always been my intent to share my data openly, as I have.

A case of two coronary scans: Nick Norwitz (high-fat, keto 🍳); Simon Hill (plant-based 🌱) First, for context, this post is largely in response to an incredibly nuanced and non-dogmatic podcast conversation between Dave Feldman and Darius Sharpe, one that had me scratching my chin and holding my ribs throughout. I definitely recommend giving it a listen. 12/10 Episode! They spend a substantial amount of time discussing cardiovascular risk and, in particular, the contrast between my coronary CT scan, showing 0 calcified and 0 soft plaque (total plaque 0mm3) despite astronomical LDL levels in the high 500s for seven years vs that of plant-based nutrition influencer Simon Hill, which showed 61.3 mm3 of plaque despite his never having had high LDL and, for the last decade or so, targeting exceptionally low LDL and ApoB, including by taking a PCSK9 inhibitor and eating a low-saturated-fat, plant-forward diet. First and foremost, there’s only so much you can conclude from a comparison of two cases. That’s obvious, so let me state that clearly. However, considering the broader context and expressed public interested, it’s still an interesting thought puzzle: What is “protecting” me from atherosclerosis, or what is leading Simon to develop plaque despite his low LDL and ApoB? I’ll address the common hand-wavy answer up front: “Well, maybe Nick has some special protective genetics.” What I’d say to this is that, although I can’t rule it out definitively since we haven’t annotated the entire human genome, my father had 99% occlusion of his left anterior descending artery at 44, and I also inherited his high Lp(a), which sits between 150 and 200. That is in addition to my historically exceptionally high LDL and a history of inflammatory bowel disease, which should further increase my risk of progression. Yet despite all of that, I have no measurable plaque, even though, as the patient, I can tell you that many cardiologists assured me I would. Additionally, the excuse that “he’s just young” doesn’t fly, since Simon is also young. We’re both in our 30s. And, as a historical comparator, children with homozygous familial hypercholesterolemia can develop plaque within a few years of being born. So, puzzle me this: Why am I not developing plaque if the relationship between LDL/ApoB exposure and plaque progression is so "remarkably consistent?" Hand-wave if you want, but I’d like something more specific please. (And I’d like anybody to tell me why they think I was confident enough to spend seven years on a self-experiment, risking my heart on the bet that this would effectively be the outcome. Was I just cocky and lucky?) That’s mostly what I want to post: a thought question to consider based on two individuals who have chosen to make their data publicly available. However, since this field is so wrapped in controversy, let me elaborate on a few more points that are related. 1. KETO-CTA. To anyone who wants to pivot and try to point the finger at KETO-CTA, I will remind you that, in 4 separate analyses, there was no relationship between LDL or ApoB and plaque progression. That remains true to this day. In fact, it's a result that remains more robust for having had our hands forced to run the additional analyses. With respect to our retraction of the April 7th Cleerly paper, which continues to be misrepresented (at this point, I have to assume willfully) I will remind you: i) we the authors, pushed forward the retraction of the April 7 paper; it was not forced upon us ii) we did so because of the Cleerly anomalies that came to light after publication – nb: Dave and I were blinded to elements of data willfully to protect the integrity of the project (there’s an irony for you) iii) we sought retraction only after company’s refusal to repeat a properly blinded rescan and quality assessment, and subsequent analyses that made the only reasonable interpretation that Cleerly systematically overestimated plaque progression iv) in none of the analyses did ApoB or LDL predict plaque progression, which was the main novel finding and remains robust. 2. Ezetimibe I am now taking ezetimibe. Critics have tried to cast this - incorrectly - as some form of me hedging, but I intentionally did not start it until after my 7-year scan, and I am not withholding any information from or about that scan. The reason I’m taking ezetimibe, which I have very clearly disclosed, including in my newsletter and in a video, is 2024 Aging Biology research suggesting cholesterol-independent effects on neuroprotection, which is something that’s important to me. The fact that I am taking it, I would cast as a testament that I’m not dogmatic on this issue, and I’m not trying to “LDL-maxx” for some aware. Rather, I’m being open, honest and transparent in my journey as a person and patient trying to make the best decision for myself at ever step of the journey. Also, on the topic of ezetimibe, see recent work by Adrian Soto-Mota DoctorTro et al., which adds another layer of nuance to LDL and ApoB management in the context of low-carbohydrate diets. TL;DR, if someone sees their LDL and ApoB rise on a low-carbohydrate / ketogenic diet, ezetimibe is likely to punch above its weight for LDL and ApoB reduction. Setting aside dogma, understanding individual biology and how to manipulate it can often be very useful in a clinical setting. I will always put data first. Acknowledging this fact is not contradictory to anything I’ve said, but completely consistent with it. It is all about individual context. Now, everyone should go listen to episode 43 of TFP_ with Dave Feldman and Darius Sharpe (Spotify, Apple and YouTube). It’s stupendous! P.S. I want to thank Dave for sharing on the podcast that, before each of my cardiac scans, I said a priori that I would release the results, whatever they were. You all know Dave to be impeccably honest. That is not in dispute, and it has always been my intent to share my data openly, as I have.

Nick Norwitz MD PhD

21,834 görüntüleme • 7 gün önce

$Yesterday at Brown University ICERM's workshop on “Agentic Scientific Computing and Scientific Machine Learning” I spoke about “Adaptive Swarms Across Scales”, making the case for scientific AI as systems that can create representations, stress them, fracture them, and enlarge the category in which future representations live. The category here is a composable and breakable working universe of science: data, hypotheses, simulations, measurements, tools, failures, figures, papers, provenance, and the transformations that connect them. Discovery happens when those transformations become executable, inspectable, composable, and capable of changing the world model they operate within. Atomistic modeling gives one category - states, forces, trajectories, observables, boundary conditions, conservation laws. Neural surrogates learn fast morphisms inside or between such categories. But discovery is higher-order: it changes which objects and morphisms are available in the first place: what variables exist, what operations are allowed, what evidence counts, what scale is active, what invariant is being preserved, and what kind of explanation the system is even capable of forming. This is scientific method as adaptive architecture: compression, stress, fracture, recomposition. Fracture matters here because it makes the logic physical: a non-commuting diagram realized in matter. The imposed load, material hierarchy, defect field, and assumed continuum description no longer map cleanly into the observed outcome. The crack is the obstruction and it identifies where the old morphism failed and where a new representation must be introduced. The physical crack and the categorical obstruction are the same event viewed in different substrates. ScienceClaw × Infinite is a machine for constructing and transforming a category of scientific artifacts. Each artifact is typed. Each operation has lineage. Each failed branch remains in the category as reusable structure. The “paper” is no longer the terminal object of science; it is one projection of a larger compositional trace, and it can be generated at any time for consumption by a human or an AI. With that the unit of scientific labor is changing. For most of the twentieth century the unit was the result (a measurement, a theorem, a synthesized molecule). It is now becoming the algorithm that produces results, and after that, the substrate of discovery itself. The static PDF is the wrong terminal object for this regime, and the role of the scientist with it. We now design algorithms that build algorithms, and eventually substrates in which such algorithms compose themselves. At that point, the scientist is no longer outside the discovery system. The scientist becomes one of the representations the system can transform. In that sense, the systems will eventually do science to us, and that is the structural consequence of the principle they are built on.$

Yesterday at Brown University ICERM's workshop on “Agentic Scientific Computing and Scientific Machine Learning” I spoke about “Adaptive Swarms Across Scales”, making the case for scientific AI as systems that can create representations, stress them, fracture them, and enlarge the category in which future representations live. The category here is a composable and breakable working universe of science: data, hypotheses, simulations, measurements, tools, failures, figures, papers, provenance, and the transformations that connect them. Discovery happens when those transformations become executable, inspectable, composable, and capable of changing the world model they operate within. Atomistic modeling gives one category - states, forces, trajectories, observables, boundary conditions, conservation laws. Neural surrogates learn fast morphisms inside or between such categories. But discovery is higher-order: it changes which objects and morphisms are available in the first place: what variables exist, what operations are allowed, what evidence counts, what scale is active, what invariant is being preserved, and what kind of explanation the system is even capable of forming. This is scientific method as adaptive architecture: compression, stress, fracture, recomposition. Fracture matters here because it makes the logic physical: a non-commuting diagram realized in matter. The imposed load, material hierarchy, defect field, and assumed continuum description no longer map cleanly into the observed outcome. The crack is the obstruction and it identifies where the old morphism failed and where a new representation must be introduced. The physical crack and the categorical obstruction are the same event viewed in different substrates. ScienceClaw × Infinite is a machine for constructing and transforming a category of scientific artifacts. Each artifact is typed. Each operation has lineage. Each failed branch remains in the category as reusable structure. The “paper” is no longer the terminal object of science; it is one projection of a larger compositional trace, and it can be generated at any time for consumption by a human or an AI. With that the unit of scientific labor is changing. For most of the twentieth century the unit was the result (a measurement, a theorem, a synthesized molecule). It is now becoming the algorithm that produces results, and after that, the substrate of discovery itself. The static PDF is the wrong terminal object for this regime, and the role of the scientist with it. We now design algorithms that build algorithms, and eventually substrates in which such algorithms compose themselves. At that point, the scientist is no longer outside the discovery system. The scientist becomes one of the representations the system can transform. In that sense, the systems will eventually do science to us, and that is the structural consequence of the principle they are built on.

Markus J. Buehler

10,095 görüntüleme • 4 ay önce

$I designed the WLFI governance vote. You may remember me. Last month I built the freeze function. The one where a single anonymous wallet can lock any token holder's assets at any time for any reason without notice or appeal. Justin Sun called it a backdoor. We called it compliance. We sued him. He sued us back. One billion dollars. His lawyers filed on April 22nd. That was phase one. Individual control. One wallet. One victim. One freeze. Phase two is collective. I needed a mechanism that would extract consent from 18,000 holders simultaneously, without anyone afterward claiming they didn't agree, without anyone pointing to a single moment when force was applied, and without anyone identifying a perpetrator, because the perpetrator would be the architecture itself. The legal team said this was important. As mentioned, I've already designed it. They asked what I called it. I said governance. We submitted 62.3 billion tokens to a ballot. The proposal: release all vesting schedules. Early supporters receive their 17 billion on a two-year cliff, with a two-year vest. Founders receive their 45.2 billion on a two-year cliff plus three-year vest with a ten percent burn. The balloting mechanism is elegant. If you accept, your holdings will be released on the published schedule. If you decline, your holdings remain frozen. Indefinitely. No timeline. No appeals process. No alternative proposal. No counteroffer. No exit. I presented this to the governance committee. Three people. All founders. They approved it in eleven minutes. I timed it because I was curious. Eleven minutes to design consent for sixty-two billion tokens. That's due diligence. 99.5% accepted. I am told this represents overwhelming community consensus. I designed the mechanism where declining means your money stays frozen forever. I am told the 99.5% approval rate proves the community supports us. Those are both true. They are also the same sentence. I put both in the press release. Four wallets controlled 40% of the total ballot. One address held 13%. Quorum required one billion. The largest participant exceeded quorum alone. We set the threshold. We also hold the addresses. The token was $0.23 in January. It trades at eight cents. Sixty-five percent decline. The investors voted to unlock assets worth one-third of what they paid. But they voted yes because the alternative was those assets staying locked forever at one-third of what they paid. I designed both options. One is loss. The other is permanent loss. They chose loss. That's participation. On the governance forum, one holder wrote: "There is no democracy. The system is a joke." Another wrote: "I'm going to put these bastards in jail." A third posted a single word: "WTF." All three accepted the proposal. I verified their wallet signatures personally. The one who promised jail voted yes fourteen minutes after his post. I have the timestamp. I keep all the timestamps. We burned 4.5 billion from the founder pool. Ten percent of our allocation. The press release said meaningful sacrifice. The communications team wanted unprecedented sacrifice. I suggested meaningful. Unprecedented implies it won't happen again. Meaningful implies nothing. Our remaining allocation after the burn. Forty point seven billion at eight cents. Three point two billion dollars. We sacrificed $360 million in locked, unsellable supply. We retained $3.2 billion that now releases on schedule. The ratio of sacrifice to retention is 1:9. I call that generosity. The press release called it alignment with the community. The community had no choice but to align back. That's sacrifice. The investors paid between $0.015 and $0.05 per token. At eight cents, some are technically in profit, sixty to four hundred percent above their entry, and they won't file lawsuits because you don't sue when you're up and because the legal costs would exceed their holdings and because by the time discovery begins the token will trade at fractions of a cent and there will be nothing to recover from anyone. They will also sell the moment their tokens unlock. All of them. Simultaneously. Which will push the price below five cents. Which means nobody is in profit. Which means nobody files lawsuits. Because there is nothing left to recover. I designed that sequence too. That's vesting. Phase one takes one wallet at a time. Phase two captures 18,000 addresses simultaneously, each of them clicking yes on the identical ballot under the identical terms I wrote, in language simple enough for a compliance officer to approve and opaque enough for a retail buyer to mistake for democracy. Same coercion. Different magnitude. Both listed under governance on the project website. Sun's billion-dollar complaint characterizes the freeze function as "a unilateral deprivation of property rights." The ballot proves otherwise. It was not unilateral. We asked. Ninety-nine point five percent said yes. Under the specific condition that saying no meant keeping nothing. That's consensus.$

I designed the WLFI governance vote. You may remember me. Last month I built the freeze function. The one where a single anonymous wallet can lock any token holder's assets at any time for any reason without notice or appeal. Justin Sun called it a backdoor. We called it compliance. We sued him. He sued us back. One billion dollars. His lawyers filed on April 22nd. That was phase one. Individual control. One wallet. One victim. One freeze. Phase two is collective. I needed a mechanism that would extract consent from 18,000 holders simultaneously, without anyone afterward claiming they didn't agree, without anyone pointing to a single moment when force was applied, and without anyone identifying a perpetrator, because the perpetrator would be the architecture itself. The legal team said this was important. As mentioned, I've already designed it. They asked what I called it. I said governance. We submitted 62.3 billion tokens to a ballot. The proposal: release all vesting schedules. Early supporters receive their 17 billion on a two-year cliff, with a two-year vest. Founders receive their 45.2 billion on a two-year cliff plus three-year vest with a ten percent burn. The balloting mechanism is elegant. If you accept, your holdings will be released on the published schedule. If you decline, your holdings remain frozen. Indefinitely. No timeline. No appeals process. No alternative proposal. No counteroffer. No exit. I presented this to the governance committee. Three people. All founders. They approved it in eleven minutes. I timed it because I was curious. Eleven minutes to design consent for sixty-two billion tokens. That's due diligence. 99.5% accepted. I am told this represents overwhelming community consensus. I designed the mechanism where declining means your money stays frozen forever. I am told the 99.5% approval rate proves the community supports us. Those are both true. They are also the same sentence. I put both in the press release. Four wallets controlled 40% of the total ballot. One address held 13%. Quorum required one billion. The largest participant exceeded quorum alone. We set the threshold. We also hold the addresses. The token was $0.23 in January. It trades at eight cents. Sixty-five percent decline. The investors voted to unlock assets worth one-third of what they paid. But they voted yes because the alternative was those assets staying locked forever at one-third of what they paid. I designed both options. One is loss. The other is permanent loss. They chose loss. That's participation. On the governance forum, one holder wrote: "There is no democracy. The system is a joke." Another wrote: "I'm going to put these bastards in jail." A third posted a single word: "WTF." All three accepted the proposal. I verified their wallet signatures personally. The one who promised jail voted yes fourteen minutes after his post. I have the timestamp. I keep all the timestamps. We burned 4.5 billion from the founder pool. Ten percent of our allocation. The press release said meaningful sacrifice. The communications team wanted unprecedented sacrifice. I suggested meaningful. Unprecedented implies it won't happen again. Meaningful implies nothing. Our remaining allocation after the burn. Forty point seven billion at eight cents. Three point two billion dollars. We sacrificed $360 million in locked, unsellable supply. We retained $3.2 billion that now releases on schedule. The ratio of sacrifice to retention is 1:9. I call that generosity. The press release called it alignment with the community. The community had no choice but to align back. That's sacrifice. The investors paid between $0.015 and $0.05 per token. At eight cents, some are technically in profit, sixty to four hundred percent above their entry, and they won't file lawsuits because you don't sue when you're up and because the legal costs would exceed their holdings and because by the time discovery begins the token will trade at fractions of a cent and there will be nothing to recover from anyone. They will also sell the moment their tokens unlock. All of them. Simultaneously. Which will push the price below five cents. Which means nobody is in profit. Which means nobody files lawsuits. Because there is nothing left to recover. I designed that sequence too. That's vesting. Phase one takes one wallet at a time. Phase two captures 18,000 addresses simultaneously, each of them clicking yes on the identical ballot under the identical terms I wrote, in language simple enough for a compliance officer to approve and opaque enough for a retail buyer to mistake for democracy. Same coercion. Different magnitude. Both listed under governance on the project website. Sun's billion-dollar complaint characterizes the freeze function as "a unilateral deprivation of property rights." The ballot proves otherwise. It was not unilateral. We asked. Ninety-nine point five percent said yes. Under the specific condition that saying no meant keeping nothing. That's consensus.

Peter Girnus 🦅

32,648 görüntüleme • 4 ay önce

Release: LichtFeld Studio v0.5.3 is out! With 316 commits merged into master, this release is a huge step forward for LichtFeld Studio. What's new in v0.5.3 • Vulkan viewer/rendering migration: New Vulkan viewport pipeline, pass graph, VkSplat renderer, Vulkan point-cloud renderer, 3DGUT/VkSplat support, improved alpha/depth composition, tighter CUDA/Vulkan interoperability, and device matching on multi-GPU systems. • RAD + LOD workflow: Added RAD file export/import, RAD LOD viewer, Spark-style GPU LOD selection, GPU-driven page prefetching, a bounded VRAM pool, out-of-core PLY-to-RAD LOD conversion, and RAD import/export speedups of approximately 3–5×. • HiGS / macro-tile inference: Added a macro-tile inference path for the Vulkan viewer, including macro sorting, batched rasterization, composition, and capacity management. • Asset Manager: Added and significantly enhanced the Asset Manager with thumbnails, SH information, faster synchronization, import-from-URL support, docked mode, data-loading popup integration, and general UI cleanup. • Viewport export: Integrated viewport export directly into the application as a toolbar/overlay tool, added fast render_view_u8-style readback paths, fixed high-resolution clipping issues, improved orthographic export parity, resolved 32K image/video export problems, and added post-export GPU resource cleanup. • Selection and tooling: Added and reworked selection toolbar controls, the Select menu, ring selection, color eyedropper, distance-from-center selection, faster point-cloud and zoomed-out selection paths, Vulkan measurement tool fixes, and drag-and-drop scene graph improvements. • UI/RmlUi platform work: Major RmlUi redesign efforts, hot reloading for RML/RCSS/Python UI files, reactive UI/store integration, viewport toolbar flyouts, improved histogram interactions, input settings enhancements, custom TRS gizmos, and numerous panel, tooltip, and localization fixes. • Windowing and UX: Added borderless window support, title bar drag/maximize/restore behavior, work-area-aware maximize functionality, resize responsiveness and performance improvements, and DPI/UI scaling fixes. • Training and data features: Added adaptive depth loss and depth gradients for the EWA rasterizer, mask loading/application fixes, a new combined Ignore+Segment mask mode, --add-splat, --freeze, improved checkpoint and training state handling, and training speed and VRAM optimizations. • COLMAP/equirectangular support: Added SPHERICAL/equirectangular camera model support and canonical EQUIRECTANGULAR handling, along with fixes for undistortion and camera export. This release will be available to all supporters as a Windows binary via approximately in about an hour. At the same time, LichtFeld Studio remains committed to being free and open source under GPLv3 and can also be built directly from source. Please consider supporting the ongoing development of LichtFeld Studio through a donation via the portal or the supporters page. Thank you to everyone who supports this project financially, contributes code, reports bugs, provides datasets, helps with the website, and contributes in countless other ways. A special thank you to our foundational sponsor Core11 and our Gold Sponsor Volinga, whose support has helped make the current state of the software possible. Thank you as well to every donor and to all of our new Bronze Sponsors. Looking ahead to v0.6 For the next major release, work will focus primarily on stability and user experience. This includes improved cleanup workflows and the ability to modify training parameters while training is in progress. I would also like to introduce a native .licht project format that allows users to save and restore their complete editor state. You can find links to our main sponsors below. Please also visit our website to discover all our Bronze Sponsors. Hint: We do not yet have a Silver Sponsor or Platinum 😉

Release: LichtFeld Studio v0.5.3 is out! With 316 commits merged into master, this release is a huge step forward for LichtFeld Studio. What's new in v0.5.3 • Vulkan viewer/rendering migration: New Vulkan viewport pipeline, pass graph, VkSplat renderer, Vulkan point-cloud renderer, 3DGUT/VkSplat support, improved alpha/depth composition, tighter CUDA/Vulkan interoperability, and device matching on multi-GPU systems. • RAD + LOD workflow: Added RAD file export/import, RAD LOD viewer, Spark-style GPU LOD selection, GPU-driven page prefetching, a bounded VRAM pool, out-of-core PLY-to-RAD LOD conversion, and RAD import/export speedups of approximately 3–5×. • HiGS / macro-tile inference: Added a macro-tile inference path for the Vulkan viewer, including macro sorting, batched rasterization, composition, and capacity management. • Asset Manager: Added and significantly enhanced the Asset Manager with thumbnails, SH information, faster synchronization, import-from-URL support, docked mode, data-loading popup integration, and general UI cleanup. • Viewport export: Integrated viewport export directly into the application as a toolbar/overlay tool, added fast render_view_u8-style readback paths, fixed high-resolution clipping issues, improved orthographic export parity, resolved 32K image/video export problems, and added post-export GPU resource cleanup. • Selection and tooling: Added and reworked selection toolbar controls, the Select menu, ring selection, color eyedropper, distance-from-center selection, faster point-cloud and zoomed-out selection paths, Vulkan measurement tool fixes, and drag-and-drop scene graph improvements. • UI/RmlUi platform work: Major RmlUi redesign efforts, hot reloading for RML/RCSS/Python UI files, reactive UI/store integration, viewport toolbar flyouts, improved histogram interactions, input settings enhancements, custom TRS gizmos, and numerous panel, tooltip, and localization fixes. • Windowing and UX: Added borderless window support, title bar drag/maximize/restore behavior, work-area-aware maximize functionality, resize responsiveness and performance improvements, and DPI/UI scaling fixes. • Training and data features: Added adaptive depth loss and depth gradients for the EWA rasterizer, mask loading/application fixes, a new combined Ignore+Segment mask mode, --add-splat, --freeze, improved checkpoint and training state handling, and training speed and VRAM optimizations. • COLMAP/equirectangular support: Added SPHERICAL/equirectangular camera model support and canonical EQUIRECTANGULAR handling, along with fixes for undistortion and camera export. This release will be available to all supporters as a Windows binary via approximately in about an hour. At the same time, LichtFeld Studio remains committed to being free and open source under GPLv3 and can also be built directly from source. Please consider supporting the ongoing development of LichtFeld Studio through a donation via the portal or the supporters page. Thank you to everyone who supports this project financially, contributes code, reports bugs, provides datasets, helps with the website, and contributes in countless other ways. A special thank you to our foundational sponsor Core11 and our Gold Sponsor Volinga, whose support has helped make the current state of the software possible. Thank you as well to every donor and to all of our new Bronze Sponsors. Looking ahead to v0.6 For the next major release, work will focus primarily on stability and user experience. This includes improved cleanup workflows and the ability to modify training parameters while training is in progress. I would also like to introduce a native .licht project format that allows users to save and restore their complete editor state. You can find links to our main sponsors below. Please also visit our website to discover all our Bronze Sponsors. Hint: We do not yet have a Silver Sponsor or Platinum 😉

MrNeRF

26,219 görüntüleme • 2 ay önce

🚨 Beijing Rolled Out the Red Carpet for Trump AND Putin in 6 Days. Its Own Investors Just Rolled Out the Exits — ¥2 Trillion Gone. ¥2 Trillion, that's ¥2,000,000,000,000. Twelve zeros. More than the entire annual GDP of Saudi Arabia. Erased in one trading session. Six days ago President Donald Trump left Beijing on Air Force One. Yesterday (May 20, 2026) Vladimir Putin walked down a red carpet into the Great Hall of the People. Today — May 21, 2026 — Chinese investors did something Beijing's propaganda machine cannot spin: they sold. An estimated ¥2 trillion (≈ US$280 billion) in market value was erased from mainland Chinese equities. The Shanghai Composite slid 2.04% and the Shenzhen Component tumbled 2.07% — both three-week lows. Hong Kong's Hang Seng closed down roughly 1%. The names that bled the hardest are the very ones Xi has been parading as proof of "tech self-reliance": Cambricon -3.19%, Zhongji Innolight -4.21%, Eoptolink -3.74%, Huagong Tech -5.79%. Even CITIC Securities — a mainland brokerage, not a foreign sceptic — noted that the pullback dates from May 14. That is the day Donald Trump landed in Beijing. This is what the market thinks of the past two weeks of choreography. The Trump Summit Beijing Sold as a Triumph The Trump–Xi summit (May 14–15) was a state-visit spectacle: military honor guards, a banquet at the Great Hall, a personal welcome from Xi. The substance was thinner. Atlantic Council's verdict: a big show with little to show for it. CNN's politics desk was more clinical: nebulous agreements on agricultural purchases, tepid commitments on oil, no firm deal to reopen the Strait of Hormuz. Trump himself said tariffs didn't even come up. Al Jazeera noted something rarer — the two sides released readouts that disagreed on what was actually agreed. The morning after the summit, US stock futures sold off across the board. Investors voted before the pundits did. Beijing's framing: historic visit. The tape's framing: priced in, sold off. The Putin Summit Beijing Sold as Strength One day before today's selloff, Xi gave Putin a red-carpet welcome — their second meeting in under a year. The two leaders presided over a sweeping signing ceremony covering trade, technology, nuclear energy and media cooperation. Xi called the relationship the "highest level in history." A joint statement took aim at Trump's planned "Golden Dome" missile shield. Optics: an axis. Reality: Putin came to Beijing with one big ask — locking in the long-stalled Power of Siberia 2 gas pipeline, the project Moscow needs to replace gas sales lost to Europe — and left without it. The Washington Post's headline was blunt: "Putin fails to secure Xi's approval for Power of Siberia 2." Price, financing and timing all remain unresolved, with Beijing reportedly holding out for prices roughly half of what Moscow wanted. Even the marquee deliverable didn't deliver. Why the Tape Doesn't Believe the Narrative Mainland investors aren't watching CCTV. They're watching the data. China just emerged from the longest stretch of producer-price deflation in decades — 41 consecutive months from October 2022 through this past February. The streak only broke in April, and not because demand came back. It broke because the Iran war pushed energy prices higher. That is imported inflation, not organic recovery. Strip out energy and the demand picture remains thin. Goldman Sachs says the property crisis is in its fourth year and not yet at a bottom. Chinese exports to the United States fell nearly 29% year-on-year in November. Youth unemployment officially stood at 16.3% in April; independent analysts argue the real figure is materially higher. Private investment remains weak — Chinese firms aren't short of liquidity, they're cautious on returns, on enforcement consistency, on whether the demand will be there tomorrow. This is the macro that propaganda cannot photoshop. The Neighbourhood: A Quiet Encirclement Look at Asia's tape today against Shanghai's. Tokyo's Nikkei rallied more than 3%, within striking distance of an all-time high set just last week. Seoul's Kospi exploded 8.42% higher — its largest single-session point gain on record, led by Samsung and SK Hynix. In Manila, "Balikatan 2026" just concluded with Japanese combat troops participating in the largest US-Philippines drills for the first time ever. Washington's Indo-Pacific lattice — AUKUS, the Quad, the trilateral US–Japan–Philippines and US–Japan–Korea formats — the architecture Beijing labels an "Asian NATO" — continues to thicken. In Brussels, Commission President Ursula von der Leyen has tied future EU-China relations explicitly to how Beijing handles Russia's war on Ukraine. And Xi is reportedly preparing his first visit to North Korea in seven years — a tell about which axis Beijing is doubling down on. Tokyo up. Seoul at a record. Shanghai down. That is not a coincidence. That is a verdict on which side of the new geopolitical fault line global capital believes will compound. Two Trillion Yuan Do Not Lie You cannot propaganda your way past a price chart. State media can stage the Trump welcome as triumph and the Putin embrace as solidarity, but the people who actually have skin in the game — Chinese savers, Chinese funds, the foreign capital still inside the wall — sold into both stories. ¥2 trillion in a single session is not a technical wobble. It is a referendum. The Trump–Xi–Putin theatre is over. The bill is being presented. And Beijing's available responses — tighter capital controls, more "national team" buying, more margin tightening, or a sharper turn toward Moscow and Pyongyang — none of them rebuild confidence. They only manage the optics of its absence. What gets priced in next? Capital controls? A managed devaluation? Another "national team" rescue? Or does the next leg down arrive before the response does? Original article by me Aric Chen. Views are my own — welcome to discuss!

🚨 Beijing Rolled Out the Red Carpet for Trump AND Putin in 6 Days. Its Own Investors Just Rolled Out the Exits — ¥2 Trillion Gone. ¥2 Trillion, that's ¥2,000,000,000,000. Twelve zeros. More than the entire annual GDP of Saudi Arabia. Erased in one trading session. Six days ago President Donald Trump left Beijing on Air Force One. Yesterday (May 20, 2026) Vladimir Putin walked down a red carpet into the Great Hall of the People. Today — May 21, 2026 — Chinese investors did something Beijing's propaganda machine cannot spin: they sold. An estimated ¥2 trillion (≈ US$280 billion) in market value was erased from mainland Chinese equities. The Shanghai Composite slid 2.04% and the Shenzhen Component tumbled 2.07% — both three-week lows. Hong Kong's Hang Seng closed down roughly 1%. The names that bled the hardest are the very ones Xi has been parading as proof of "tech self-reliance": Cambricon -3.19%, Zhongji Innolight -4.21%, Eoptolink -3.74%, Huagong Tech -5.79%. Even CITIC Securities — a mainland brokerage, not a foreign sceptic — noted that the pullback dates from May 14. That is the day Donald Trump landed in Beijing. This is what the market thinks of the past two weeks of choreography. The Trump Summit Beijing Sold as a Triumph The Trump–Xi summit (May 14–15) was a state-visit spectacle: military honor guards, a banquet at the Great Hall, a personal welcome from Xi. The substance was thinner. Atlantic Council's verdict: a big show with little to show for it. CNN's politics desk was more clinical: nebulous agreements on agricultural purchases, tepid commitments on oil, no firm deal to reopen the Strait of Hormuz. Trump himself said tariffs didn't even come up. Al Jazeera noted something rarer — the two sides released readouts that disagreed on what was actually agreed. The morning after the summit, US stock futures sold off across the board. Investors voted before the pundits did. Beijing's framing: historic visit. The tape's framing: priced in, sold off. The Putin Summit Beijing Sold as Strength One day before today's selloff, Xi gave Putin a red-carpet welcome — their second meeting in under a year. The two leaders presided over a sweeping signing ceremony covering trade, technology, nuclear energy and media cooperation. Xi called the relationship the "highest level in history." A joint statement took aim at Trump's planned "Golden Dome" missile shield. Optics: an axis. Reality: Putin came to Beijing with one big ask — locking in the long-stalled Power of Siberia 2 gas pipeline, the project Moscow needs to replace gas sales lost to Europe — and left without it. The Washington Post's headline was blunt: "Putin fails to secure Xi's approval for Power of Siberia 2." Price, financing and timing all remain unresolved, with Beijing reportedly holding out for prices roughly half of what Moscow wanted. Even the marquee deliverable didn't deliver. Why the Tape Doesn't Believe the Narrative Mainland investors aren't watching CCTV. They're watching the data. China just emerged from the longest stretch of producer-price deflation in decades — 41 consecutive months from October 2022 through this past February. The streak only broke in April, and not because demand came back. It broke because the Iran war pushed energy prices higher. That is imported inflation, not organic recovery. Strip out energy and the demand picture remains thin. Goldman Sachs says the property crisis is in its fourth year and not yet at a bottom. Chinese exports to the United States fell nearly 29% year-on-year in November. Youth unemployment officially stood at 16.3% in April; independent analysts argue the real figure is materially higher. Private investment remains weak — Chinese firms aren't short of liquidity, they're cautious on returns, on enforcement consistency, on whether the demand will be there tomorrow. This is the macro that propaganda cannot photoshop. The Neighbourhood: A Quiet Encirclement Look at Asia's tape today against Shanghai's. Tokyo's Nikkei rallied more than 3%, within striking distance of an all-time high set just last week. Seoul's Kospi exploded 8.42% higher — its largest single-session point gain on record, led by Samsung and SK Hynix. In Manila, "Balikatan 2026" just concluded with Japanese combat troops participating in the largest US-Philippines drills for the first time ever. Washington's Indo-Pacific lattice — AUKUS, the Quad, the trilateral US–Japan–Philippines and US–Japan–Korea formats — the architecture Beijing labels an "Asian NATO" — continues to thicken. In Brussels, Commission President Ursula von der Leyen has tied future EU-China relations explicitly to how Beijing handles Russia's war on Ukraine. And Xi is reportedly preparing his first visit to North Korea in seven years — a tell about which axis Beijing is doubling down on. Tokyo up. Seoul at a record. Shanghai down. That is not a coincidence. That is a verdict on which side of the new geopolitical fault line global capital believes will compound. Two Trillion Yuan Do Not Lie You cannot propaganda your way past a price chart. State media can stage the Trump welcome as triumph and the Putin embrace as solidarity, but the people who actually have skin in the game — Chinese savers, Chinese funds, the foreign capital still inside the wall — sold into both stories. ¥2 trillion in a single session is not a technical wobble. It is a referendum. The Trump–Xi–Putin theatre is over. The bill is being presented. And Beijing's available responses — tighter capital controls, more "national team" buying, more margin tightening, or a sharper turn toward Moscow and Pyongyang — none of them rebuild confidence. They only manage the optics of its absence. What gets priced in next? Capital controls? A managed devaluation? Another "national team" rescue? Or does the next leg down arrive before the response does? Original article by me Aric Chen. Views are my own — welcome to discuss!

Aric Chen

125,716 görüntüleme • 3 ay önce

‡ Brant – Fast, Expensive, and Worrying In a recent 2yo MSW race at Santa Anita, Brant, a $3m OBS March sales purchase, made a winning debut for his high-profile connections. Sent off at odds of 4/5, he tracked the pace before taking command on the turn, and drew off to win by 5 1/4 lengths. He recorded a very fast raw time, and a 101 Beyer figure. His high auction price was due largely to his having breezed an eighth of a mile in :09 3/5 at OBS. In the wake of the bidding, Amr Zedan, who purchased the colt on the recommendation of his trainer, Bob Baffert, and bloodstock agent Donato Lanni, was quoted in TDN as saying: "These horses are difficult to come by. He ticked all the boxes. He was a very precocious Gun Runner with a great pedigree. And more importantly, if you have someone like Mr. [Bob] Baffert in your corner, that gives you the courage and the guts to just go after quality. And you'll know they are in the best hands to turn them into champions. This one is for the team: Donato and obviously Bob and his ability to turn them into champions. So if you have the great team, the great training, the rest is easy. Was he pricey? Yes. But quality dictates price. So I never hesitated.” The hyperbole, and boilerplate optimism, are understandable, as even owners with very deep pockets prefer not to dwell on their inevitably long lists of expensive failures. But Zedan and their team have also enjoyed a number of high-profile successes, including Taiba, another Gun Runner colt, which won the Santa Anita Derby and the Malibu, both Grade I, before being retired to stud. Lanni, who has signed for at least some of Zedan's other good horses, was quoted as saying: "This is what the boss [Baffert] wanted and what Amr wanted. Gun Runner is a tremendous stallion and he worked really good and galloped out good. He did everything you want one to do.” Again, boilerplate, and if one were to take the reactions of the owner and agent at face value, it would be easy to arrive at the conclusion that the horse had no faults. But that would be naïve. So, let's first take a look at Brant's pedigree. Gun Runner is a "top" sire, and well-capable of getting high-class runners. It's a bit too early to fully judge him as a source of durability, but excluding his current crop of 2yos, his runners are only averaging 10 career starts. That number will rise, but likely not enough to reach, let alone exceed the contemporary industry average of ~15. In other words, though he himself raced 19 times, and won his swan song (the Pegasus World Cup) at five, there is no evidence to suggest that he is likely to eventually prove to be a particular source of durability. Brant's dam, Tynan, raced 13 times, and he is her first registered foal. His second dam, Pappascat, has produced at least five foals to have raced, and while only one has reached 20 starts, four of the five raced at least 12 times, which by today's degraded standards, isn't so bad. The fact that the coat color inherited by Brant, and his dam, can be traced to a notably unsound influence, Unbridled's Song, may or may not be meaningful. But I wouldn't ignore it as a potentially worrying connection. Brant's dam-sire, Liam’s Map, was lightly raced himself, and both his sire (US) and dam have poor records in terms of durability. As a sire, he has thus far produced numbers similar to Gun Runner. So while Brant's first two dams promise more than many that I have come across in similar assessments, and his sire displayed durability on the track, his overall pedigree suggests average durability at best. And what about the word that Mr. Zedan used twice in the above quote – "quality"? Well, Curlin is a quality sire, but in terms of bottom-line production, it leaves a lot to be desired. That's not to say that there are no good runners to be found, in fact the closely related Pappacap, under Brant's second-dam, was a Gr. III winner, and twice Grade I placed. However, through Brant's first six dams at least, I believe there to be just a single Grade I winner, Al Qasr, a Champion stayer in Peru, which appears under his fourth-dam. That is not, by any reasonable definition, a strong bottom-line, although it is fair to say that Brant's first dam is unproven. *** What might be learned from Brant's debut race? Everyone can see that he ran fast, and was much the best of that field, but I would say that there were some nuanced aspects of his performance that were both interesting, and worrying. Those nuances relate to his action, as viewed both through the pan shot, and head-on. Watching the basic (pan) view, Brant appears to display some "knee action". It isn't extreme, but also isn't the type of action that ideally suits dirt runners, and I wouldn't say that the colt appears totally comfortable. Here's a brief, related post on knee action, for reference: Then, we have the (embedded) head-on view, in which Brant displays seriously distorted action in his near-fore (left front leg; right when viewed head-on). It's a fairly extreme example of what is called "winging", and for a number of reasons, does not bode well for durability. Such action is never desirable, and is typically related to certain foreleg conformation flaws, which may include an offset knee, and/or toeing in or out, etc. Note also that under typical American racing conditions (i.e. tightly turning, left-handed tracks), the left front leg is subjected to the greatest torque, which amplifies the potential for injury. It should go without saying that there are occasional anomalies, horses with poor action that remain sound despite such flaws. But they are the exceptions, and it is not uncommon to find abbreviated careers associated with such action. For further reference, here is a link to a closely related post that I wrote after a filly named Amor Fati broke her maiden in eye-catching style in February of 2024. She has made just one further start, and hasn't recorded a work over the past 15 months. What's interesting about Brant is that there are two separate issues relating to his action, and that the some obvious mitigating steps that could have been, or should be taken, in efforts to keep him sound, were not, and are unlikely to ever be taken. First, with regard to his knee action, keep in mind that it is a characteristic that is more typically associated with turf horses. And guess what? Brant's dam was a turf horse. His second-dam was also at her best on turf, and was Gr. II placed on that surface. Also, in some respects, he physically resembles that female line more than his sire. Of course no one who spends millions on a horse that breezed exceptionally fast at a 2yo sale would be thinking "turf", given that the commercial market, and stakes schedules greatly favor dirt runners. But at the same time, it would be a mistake to assume that a fast breeze at OBS necessarily indicates that a given horse will be best suited to dirt. I say that partly because the OBS track features an all-weather surface called "Safetrack", which is far from being identical to dirt racing surfaces. While countless horses have gone through those sales and excelled on dirt, it should never been taken for granted that it will be a preference. And to further flesh out the point, take a look at Brant's breeze video through this link (his pedigree page can also be seen): Despite being rushed to cover a furlong much faster than he ever will again, I would say that he looks smoother, and displays slightly less knee action than in his recent debut race. Why? Could it be, perhaps, because he would prove more comfortable on turf and/or synthetic tracks, than dirt surfaces? Given how he ran first out, this is purely an academic point, as there is virtually no chance that his connections would consider switching surfaces, unless his form on dirt were to deteriorate badly. The second, more important point, relates to Brant's distorted action. I can't find a head-on conformation photo of the colt, but would be very surprised if he does not display flaws in his near-fore. Even in the very unlikely event that the leg were to appear correct, such distorted action would have been on display at the sale, as well as in pre-training at Eddie Woods' farm. And it defies belief that experienced horsemen who were prepared to purchase an extremely expensive horse for an important client would not have covered that base. Which in turn begs two important questions. First, why recommend the purchase of a very expensive horse, no matter how fast, that has yet to race, and displays such distorted action? Secondly, why choose to give such a horse to a trainer whose style and history suggest that injury risk would likely be amplified, rather than mitigated? The answers to those questions help to illuminate why the breed-to-sell paradigm is so insidious. As long as breeders and stud farms are willing to overpay for stallion prospects, and turn blind eyes to conformation defects, distorted action, lack of durability, and modest female families, the connections of horses like Brant need only hope that they hold together long enough to win one or two big races, enabling them cash in. These are the number of career starts made by Zedan's previous best (and expensive) male runners: 9 Arabian Lion 6 Arabian Knight 9 Muth 10 Medina Spirit 11 Hejazi And as long as the AGSC (American Graded Stakes Committee) continues to enable the paradigm through its dishonest KY Derby "prep" Grade I designations, the ultimately damaging feedback loop will likely continue. The answer to the second question is closely entwined, and should be obvious. For all of his faults as a trainer, Baffert has long produced results in stakes races that lead to valuable stud careers, so owners who wish to play the breed-to-sell game consider him to be a logical choice. Of course Baffert and Zedan are hardly alone in supporting the dubious paradigm. But that a horse like Brant could bring millions at a sale, and be given to a trainer who trains very hard, and has compiled a very poor safety record, underscores the extreme degree to which the value of durability has been marginalized by the industry. One final note, and it's a genuine qualification that I always make when producing this type of post, I hope that Brant will enjoy a long, injury-free career. But if I were a betting man...

Tinky

94,348 görüntüleme • 1 yıl önce