正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

1-layer, 8-bits parameter neural net doing pattern completion with sparse distributed activations and grid-cell-like vector fields can learn to retopo, maybe. #b3d #blender #retopology

Jeacom

3,643 subscribers

22,379 次观看 • 3 年前 •via X (Twitter)

#b3d #blender #retopology

Anya Rossi• Live Now

Private livecam show

7 条评论

Jeacom 的头像

Jeacom3 年前

Still having trouble with the brush coordinates thing.

Daniel Kreuter 的头像

Daniel Kreuter3 年前

I love this idea! I’m looking forward to seeing what will become this concept 😊

L̟̹̩̾̐apisSea🔶 的头像

L̟̹̩̾̐apisSea🔶3 年前

In my experience retopology is not really an issue of how to place the next quad in a grid, it's more an issue of how to achieve the wanted edge flow and density in certain area while minimizing poles or triangles and maximizing the preserved detail of the base shape.

Jeacom 的头像

Jeacom3 年前

well, this is gonna be a quad brush and not a auto-retopo button so unfortunately, solving poles is a task for the user. but I guess automatically placing and connecting dozens of quads in sensible locations with a single brush stroke is going to release a bit of the burden.

AFX 的头像

AFX3 年前

This looks awesome 👏

ADecadeAtMost 的头像

ADecadeAtMost3 年前

How’s Blender sending and receiving data to the model and whete is the model running?

Jeacom 的头像

Jeacom3 年前

The model is running on CPU locally inside blender's thread. its all mostly custom un-optmized python code and some Cython for vector crunching.

相关视频

Backpropagation by hand ✍️ ~ 11 steps walkthrough below Backpropagation is the algorithm that actually trains a neural network, and it is where most people stop following along. It is not calculus you cannot do. It is matrix multiplication, working backward, one layer at a time. So I drew and calculated one entirely by hand. Goal: push the loss gradient back through a 3-layer network and land on a new value for every weight and bias. = 1. Given = A 3-layer perceptron, an input X, predictions Ypred = [0.5, 0.5, 0], and the truth Ytarget = [0, 1, 0]. = 2. Backprop gradient cells = Let us draw empty cells for every gradient we are about to compute. The shape of the answer comes first. = 3. Layer 3 softmax = We get dL/dz3 straight from Ypred minus Ytarget = [0.5, -0.5, 0]. No chain rule needed, and that shortcut is the whole reason softmax and cross-entropy are paired. = 4. Layer 3 weights and biases = Let us multiply dL/dz3 by [a2 | 1]. One multiplication gives the gradient for W3 and b3 together. = 5. Layer 2 activations = We multiply dL/dz3 by W3 to get dL/da2. The gradient moves back across a layer the same way the signal moved forward. = 6. Layer 2 ReLU = Let us pass it through the gate: keep the gradient where the activation was positive, zero it everywhere else. = 7. Layer 2 weights and biases = We multiply dL/dz2 by [a1 | 1]. The same figure as step 4, one layer up. = 8. Layer 1 activations = Let us multiply dL/dz2 by W2. = 9. Layer 1 ReLU = We apply the same gate again, now on a1. = 10. Layer 1 weights and biases = Let us multiply dL/dz1 by [x | 1], and every weight in the network now has a gradient. = 11. Update = We subtract, and the network has learned. In practice a learning rate scales this step. The gradients: dL/dz3 = [0.5, -0.5, 0] dL/da1 = [1, -2, 2, -1] dL/dz1 = [0, -2, 2, -1] The takeaway: matrix multiplication is all you need. Just like the forward pass, backpropagation is matrix multiplications end to end. You can do every one by hand, slowly and imperfectly, which is exactly why a GPU's ability to do them fast mattered so much to deep learning. 💾 Save this post!

Backpropagation by hand ✍️ ~ 11 steps walkthrough below Backpropagation is the algorithm that actually trains a neural network, and it is where most people stop following along. It is not calculus you cannot do. It is matrix multiplication, working backward, one layer at a time. So I drew and calculated one entirely by hand. Goal: push the loss gradient back through a 3-layer network and land on a new value for every weight and bias. = 1. Given = A 3-layer perceptron, an input X, predictions Ypred = [0.5, 0.5, 0], and the truth Ytarget = [0, 1, 0]. = 2. Backprop gradient cells = Let us draw empty cells for every gradient we are about to compute. The shape of the answer comes first. = 3. Layer 3 softmax = We get dL/dz3 straight from Ypred minus Ytarget = [0.5, -0.5, 0]. No chain rule needed, and that shortcut is the whole reason softmax and cross-entropy are paired. = 4. Layer 3 weights and biases = Let us multiply dL/dz3 by [a2 | 1]. One multiplication gives the gradient for W3 and b3 together. = 5. Layer 2 activations = We multiply dL/dz3 by W3 to get dL/da2. The gradient moves back across a layer the same way the signal moved forward. = 6. Layer 2 ReLU = Let us pass it through the gate: keep the gradient where the activation was positive, zero it everywhere else. = 7. Layer 2 weights and biases = We multiply dL/dz2 by [a1 | 1]. The same figure as step 4, one layer up. = 8. Layer 1 activations = Let us multiply dL/dz2 by W2. = 9. Layer 1 ReLU = We apply the same gate again, now on a1. = 10. Layer 1 weights and biases = Let us multiply dL/dz1 by [x | 1], and every weight in the network now has a gradient. = 11. Update = We subtract, and the network has learned. In practice a learning rate scales this step. The gradients: dL/dz3 = [0.5, -0.5, 0] dL/da1 = [1, -2, 2, -1] dL/dz1 = [0, -2, 2, -1] The takeaway: matrix multiplication is all you need. Just like the forward pass, backpropagation is matrix multiplications end to end. You can do every one by hand, slowly and imperfectly, which is exactly why a GPU's ability to do them fast mattered so much to deep learning. 💾 Save this post!

Tom Yeh

941,313 次观看 • 5 天前

I think I found the most intuitive Blender texturing add-on for artists who work with NPR models relying on hand-painting 🥹 This is “Paint System” and I found it really simple to use and navigate! It really feels like using PS/CSP in Blender ദ്ദി(ᵔᗜᵔ) Here’s a simple video showing how layers work (very simple clipping mask & alpha locking too), changing layer blend mode, and you can then bake the layers into 1 & export image! Opacity of each layer can also be adjusted with a slider like in PS/CSP but I forgot to show it hahah It also comes with 2 very nice brushes!! I know there are other add-ons for the same job or can do even more things but just wanted to share this since I found it the most intuitive as someone who hand paints textures a lot in other art programs and have non-existent knowledge over nodes in Blender ;-; Link to the add-on in replies and happy painting!!!! (ᗒ⩊ᗕ)⸝ި ʕᦏ⌎ (yes it’s free! with an option to donate to the creator to support them too)

I think I found the most intuitive Blender texturing add-on for artists who work with NPR models relying on hand-painting 🥹 This is “Paint System” and I found it really simple to use and navigate! It really feels like using PS/CSP in Blender ദ്ദി(ᵔᗜᵔ) Here’s a simple video showing how layers work (very simple clipping mask & alpha locking too), changing layer blend mode, and you can then bake the layers into 1 & export image! Opacity of each layer can also be adjusted with a slider like in PS/CSP but I forgot to show it hahah It also comes with 2 very nice brushes!! I know there are other add-ons for the same job or can do even more things but just wanted to share this since I found it the most intuitive as someone who hand paints textures a lot in other art programs and have non-existent knowledge over nodes in Blender ;-; Link to the add-on in replies and happy painting!!!! (ᗒ⩊ᗕ)⸝ި ʕᦏ⌎ (yes it’s free! with an option to donate to the creator to support them too)

𝗞𝗔𝗥𝗜𝗛𝗔 🌘🍀 3D Artist ☻

19,713 次观看 • 5 个月前

Google Ironwood TPU Memory Hierarchy in 9 levels by hand ✍️ 1. Bit – The most basic unit of information, the on–off decision from which every number, tensor, and model state is ultimately constructed. 2. FP8 (1×8 → 8 bits) – Eight bits are grouped to form a floating-point value, typically used for inference, where reduced precision is a deliberate trade-off to maximize throughput and efficiency. 3. BF16 (×2 → 16 bits) – Two FP8-scale chunks are combined to gain more dynamic range and stability, while still staying friendly to high-throughput hardware. 4. Tensor tile (×1024 → 1K) – Data moves through the chip in blocks of 1024 values at a time, defining the granularity at which tensors are fetched and manipulated. 5. Matrix Multiplication Unit (MXU) (×64 → 64K) – A systolic array where matrix multiplication is not abstract but physical, with tensor tiles flowing through fixed hardware to achieve the highest possible throughput. 6. Vector Memory (VMEM) (×2048 → 128M) – On-chip working memory that holds activations, partial results, and intermediates, sized specifically to keep the systolic array busy without stalling. 7. Common Memory (CMEM) (×8 → 1 GB) – A small but critical shared memory sitting between VMEM and HBM, used for staging, accumulation, synchronization, and cross-lane coordination. 8. HBM (×96 → 96 GB) – Off-chip high-bandwidth memory where model weights and large states live, implemented as HBM3e with 16 stacks at 6 GB each, for a total of 96 GB. 9. Dual-Die (x2 → 192GB) – Two tightly coupled compute dies operate as a single logical accelerator, each with its own local HBM, effectively doubling memory capacity and bandwidth while allowing tensors and activations to stream seamlessly across dies as if they lived on one chip. I created this drawing for this week's seminar. I’ll take you through these 9 levels in a beginner-friendly way by hand ✍️. RSVP 👉

Google Ironwood TPU Memory Hierarchy in 9 levels by hand ✍️ 1. Bit – The most basic unit of information, the on–off decision from which every number, tensor, and model state is ultimately constructed. 2. FP8 (1×8 → 8 bits) – Eight bits are grouped to form a floating-point value, typically used for inference, where reduced precision is a deliberate trade-off to maximize throughput and efficiency. 3. BF16 (×2 → 16 bits) – Two FP8-scale chunks are combined to gain more dynamic range and stability, while still staying friendly to high-throughput hardware. 4. Tensor tile (×1024 → 1K) – Data moves through the chip in blocks of 1024 values at a time, defining the granularity at which tensors are fetched and manipulated. 5. Matrix Multiplication Unit (MXU) (×64 → 64K) – A systolic array where matrix multiplication is not abstract but physical, with tensor tiles flowing through fixed hardware to achieve the highest possible throughput. 6. Vector Memory (VMEM) (×2048 → 128M) – On-chip working memory that holds activations, partial results, and intermediates, sized specifically to keep the systolic array busy without stalling. 7. Common Memory (CMEM) (×8 → 1 GB) – A small but critical shared memory sitting between VMEM and HBM, used for staging, accumulation, synchronization, and cross-lane coordination. 8. HBM (×96 → 96 GB) – Off-chip high-bandwidth memory where model weights and large states live, implemented as HBM3e with 16 stacks at 6 GB each, for a total of 96 GB. 9. Dual-Die (x2 → 192GB) – Two tightly coupled compute dies operate as a single logical accelerator, each with its own local HBM, effectively doubling memory capacity and bandwidth while allowing tensors and activations to stream seamlessly across dies as if they lived on one chip. I created this drawing for this week's seminar. I’ll take you through these 9 levels in a beginner-friendly way by hand ✍️. RSVP 👉

Tom Yeh

30,489 次观看 • 6 个月前

🎓 Learn how to create a Cherry Wood Material in a matter of seconds in my latest video series (AAA) Pro Tips! ____________________________________________________ In this video, the steps are as follows: 1. Create a slightly red-colored fill layer with high gloss to act as the woods base material. 2. Add a second fill layer with a black mask, and inside that mask add a stripes greyscale set to planar projection. These stripes will act as the wood rings and should be projected onto your mesh in a natural way. The stripes should also be set to max softness, and followed by a Directional Noise 3 with a blend mode set to Difference. 3. Add a levels to tweak the mask and add an anchor point to the mask as well. 4. Create another fill layer with a dark brown diffuse, and store the information from the anchor point into its mask. Apply a levels to increase its intensity, and then add a fill layer up top with the anchorpoint information applied again, but set its blending mode to Subtract. 5. Create an additional fill layer with a very dark diffuse channel. Add a mask to the fill layer and inside its mask add a fill layer containing the anchorpoint information, from the Wood Rings. This fill layer can now be used to add some dark contrast to the center of the wood rings 6. A warp filter should also be used to apply some subtle imperfections to the wood rings and grain. Now we are left with a very nice result of Cherry wood! ____________________________________________________ More AAA Game Dev Tips can be found on my YouTube channel here: Stay tuned for more weekly Tips. Let me know below what you would like to learn in future videos! 👇 Happy Texturing! 💚 #gameart #texturing

🎓 Learn how to create a Cherry Wood Material in a matter of seconds in my latest video series (AAA) Pro Tips! In this video, the steps are as follows: 1. Create a slightly red-colored fill layer with high gloss to act as the woods base material. 2. Add a second fill layer with a black mask, and inside that mask add a stripes greyscale set to planar projection. These stripes will act as the wood rings and should be projected onto your mesh in a natural way. The stripes should also be set to max softness, and followed by a Directional Noise 3 with a blend mode set to Difference. 3. Add a levels to tweak the mask and add an anchor point to the mask as well. 4. Create another fill layer with a dark brown diffuse, and store the information from the anchor point into its mask. Apply a levels to increase its intensity, and then add a fill layer up top with the anchorpoint information applied again, but set its blending mode to Subtract. 5. Create an additional fill layer with a very dark diffuse channel. Add a mask to the fill layer and inside its mask add a fill layer containing the anchorpoint information, from the Wood Rings. This fill layer can now be used to add some dark contrast to the center of the wood rings 6. A warp filter should also be used to apply some subtle imperfections to the wood rings and grain. Now we are left with a very nice result of Cherry wood! More AAA Game Dev Tips can be found on my YouTube channel here: Stay tuned for more weekly Tips. Let me know below what you would like to learn in future videos! 👇 Happy Texturing! 💚 #gameart #texturing

Cohen Brawley

25,184 次观看 • 2 年前

Did you know that in Blender you can easily make pillows with the 'Cloth' modifier? ✨ 1. Select the object that's gonna be beneath your pillow 2. Press 'Collision' on the physics tab 3. Add a cube and add some loops to it with Ctrl+R 4. Press 'Cloth' on the physics tab 5. Enable 'Pressure' and set the custom target volume to 1. You can tweak this value depending on how big your pillow is 6. Decrease the collision distance in 'Object Collisions' 7. Press space to play the animation 8. Press Ctrl+3 to add a subdiv modifier with 3 levels to make it look smoother! You can choose a frame that you like and apply the 'Cloth' modifier to make it into a static mesh by going to the Modifiers tab, hovering over the 'Cloth' Modifier and pressing Ctrl+A! For better results add more loops to your original mesh! ✨ #b3d

Did you know that in Blender you can easily make pillows with the 'Cloth' modifier? ✨ 1. Select the object that's gonna be beneath your pillow 2. Press 'Collision' on the physics tab 3. Add a cube and add some loops to it with Ctrl+R 4. Press 'Cloth' on the physics tab 5. Enable 'Pressure' and set the custom target volume to 1. You can tweak this value depending on how big your pillow is 6. Decrease the collision distance in 'Object Collisions' 7. Press space to play the animation 8. Press Ctrl+3 to add a subdiv modifier with 3 levels to make it look smoother! You can choose a frame that you like and apply the 'Cloth' modifier to make it into a static mesh by going to the Modifiers tab, hovering over the 'Cloth' Modifier and pressing Ctrl+A! For better results add more loops to your original mesh! ✨ #b3d

passivestar

174,886 次观看 • 3 年前

The previs workflow is still leveling up! My AI filmmaking process is starting to feel less like wrangling AI and more like animation. Still just cubes in Blender, but now I’m moving the character through space and doing a handheld camera follow. Same basic process: 1. Generate a start frame in Midjourney 2. Block out the scene in Blender, animate the camera 3. Feed both to Seedance Seedance nailed the gritty handheld feel and followed the character movement around the car way better than expected. You can get way more ambitious with your directing than I thought.

The previs workflow is still leveling up! My AI filmmaking process is starting to feel less like wrangling AI and more like animation. Still just cubes in Blender, but now I’m moving the character through space and doing a handheld camera follow. Same basic process: 1. Generate a start frame in Midjourney 2. Block out the scene in Blender, animate the camera 3. Feed both to Seedance Seedance nailed the gritty handheld feel and followed the character movement around the car way better than expected. You can get way more ambitious with your directing than I thought.

Reid Hannaford

96,712 次观看 • 29 天前

How to model a beautiful stylized tornado in Blender? ✨ 1. Model an amogus 2. With proportional editing enabled, gently rotate the head of the amogus, and then move it outwards! 3. Make a couple of copies of the amogus for more detail! 4. Tweak details to achieve your creative vision! 5. Realise that it looks nothing like a tornado. More like some weird-ass stylised flame from a torch 6. Cry a little 7. Model a torch 8. Maybe add a person holding the torch, because why not... 9. Enjoy your work! Important: when you're done modeling your tornado.. I mean torch, be sure to save your work by pressing Alt+F4! 🥰 #b3d

How to model a beautiful stylized tornado in Blender? ✨ 1. Model an amogus 2. With proportional editing enabled, gently rotate the head of the amogus, and then move it outwards! 3. Make a couple of copies of the amogus for more detail! 4. Tweak details to achieve your creative vision! 5. Realise that it looks nothing like a tornado. More like some weird-ass stylised flame from a torch 6. Cry a little 7. Model a torch 8. Maybe add a person holding the torch, because why not... 9. Enjoy your work! Important: when you're done modeling your tornado.. I mean torch, be sure to save your work by pressing Alt+F4! 🥰 #b3d

passivestar

101,978 次观看 • 3 年前

you tend to hear this a lot from people outside or new to ML, and I often point to a talk Ilya gave a few years back: 1) think of any decent deep neural net that has enough memory and sequential ops as just a big parallel computer 2) training this neural net is doing search over computer programs that maximize your objective 3)unless you have some large bottleneck (and given you can successfully optimize this system) you’ll find that these parallel computers are highly robust to architectural changes. 4) this is because computers are great at simulating each other. your new architecture can usually be straightforwardly simulated ‘inside’ your old architecture. 5) it’s not that architecture doesn’t matter, but it mostly matters with respect to (1) fundamental bottlenecks in this parallel computer (2) modifications that make models easier to optimize, since this argument only holds if your optimization is good (3) compute efficiency/system efficiency wins that make learning easier or faster. 6) it’s quite possible that new architectures will lead to breakthroughs in machine learning, but we should first start with bottlenecks, not naturalist intuitions about the ‘form’ of AI should take. until you understand this it seems surprising that small models trained longer are better than undertrained big models, that depth and width are surprisingly interchangeable, that talking to a model with an MoE or sparse attention or linear attention is approximately the same iso evals.

you tend to hear this a lot from people outside or new to ML, and I often point to a talk Ilya gave a few years back: 1) think of any decent deep neural net that has enough memory and sequential ops as just a big parallel computer 2) training this neural net is doing search over computer programs that maximize your objective 3)unless you have some large bottleneck (and given you can successfully optimize this system) you’ll find that these parallel computers are highly robust to architectural changes. 4) this is because computers are great at simulating each other. your new architecture can usually be straightforwardly simulated ‘inside’ your old architecture. 5) it’s not that architecture doesn’t matter, but it mostly matters with respect to (1) fundamental bottlenecks in this parallel computer (2) modifications that make models easier to optimize, since this argument only holds if your optimization is good (3) compute efficiency/system efficiency wins that make learning easier or faster. 6) it’s quite possible that new architectures will lead to breakthroughs in machine learning, but we should first start with bottlenecks, not naturalist intuitions about the ‘form’ of AI should take. until you understand this it seems surprising that small models trained longer are better than undertrained big models, that depth and width are surprisingly interchangeable, that talking to a model with an MoE or sparse attention or linear attention is approximately the same iso evals.

will depue

215,019 次观看 • 7 个月前

How were humans able to recognize that Newton's laws of motion govern both the flight of a bird and the motion of a pendulum? This ability to identify the same mathematical patterns across vastly different contexts lies at the heart of scientific discovery—whether studying the aerodynamics of bird wings or designing the blades of a wind turbine. Yet, AI systems often struggle to discern these deep structural similarities. 💡The key may lie in mathematical isomorphisms—patterns that preserve their relationships regardless of context. For example, the same principles of fluid dynamics apply to blood flowing through arteries and air streaming over an airplane wing, or the motion of a molecule. This raises a fundamental question in artificial intelligence: how can we enable machines to understand the world through these invariant structures rather than surface features? 🚀Our work introduces Graph-Aware Isomorphic Attention, improving how Transformers recognize patterns across domains. Drawing from category theory, models can learn unifying structural principles that describe phenomena as diverse as the hierarchical assembly of spider silk proteins and the compositional patterns in music. By making these deep similarities explicit, Isomorphic Attention enables AI to reason more like humans do—seeing past surface differences to grasp fundamental patterns that unite seemingly disparate fields. Through this lens, AI systems can learn and generalize, moving beyond superficial pattern matching to true structural understanding. The implications span from scientific discovery to engineering design, offering a new approach to artificial intelligence that mirrors how humans grasp the underlying unity of natural phenomena. Some key insights include: 1️⃣ Graph Isomorphism Neural Networks (GINs): GIN-style aggregation ensures structurally distinct graphs map to distinct embeddings, improving generalization and avoiding relational pattern collapse. 2️⃣ Category Theory Perspective: Transformers as functors preserve structural relationships. Sparse-GIN refines attention into sparse adjacency matrices, unifying domain knowledge across tasks. 3️⃣ Information Bottleneck & Sparsification: Sparsity reduces overfitting by filtering irrelevant edges, aligning with natural systems. Sparse-GIN outperforms dense attention by focusing on crucial connections. 4️⃣ Hierarchical Representation Learning: GIN-Attention captures multiscale patterns, mirroring structures like spider silk. Nested GINs model local and global dependencies across fields. 5️⃣ Practical Impact: Sparse-GIN enables domain-specific fine-tuning atop pre-trained Transformer foundation models, reducing the need for full retraining. Other impacts: ✅Real-World Relevance: Whether we are looking at protein structures, designing new materials, or working on social network analytics, graph-aware Transformers can capture subtle relational patterns traditional architectures may miss. ✅The juncture of graph isomorphism theory, category theory, and sparsification, these GIN-Transformers step beyond sequential modeling to tackle the relational nature of complex data. #Transformers #GraphNeuralNetworks #AI #MachineLearning #Isomorphism #CategoryTheory #ArtificialIntelligence #DeepLearning Link to paper & code in response ⤵️

How were humans able to recognize that Newton's laws of motion govern both the flight of a bird and the motion of a pendulum? This ability to identify the same mathematical patterns across vastly different contexts lies at the heart of scientific discovery—whether studying the aerodynamics of bird wings or designing the blades of a wind turbine. Yet, AI systems often struggle to discern these deep structural similarities. 💡The key may lie in mathematical isomorphisms—patterns that preserve their relationships regardless of context. For example, the same principles of fluid dynamics apply to blood flowing through arteries and air streaming over an airplane wing, or the motion of a molecule. This raises a fundamental question in artificial intelligence: how can we enable machines to understand the world through these invariant structures rather than surface features? 🚀Our work introduces Graph-Aware Isomorphic Attention, improving how Transformers recognize patterns across domains. Drawing from category theory, models can learn unifying structural principles that describe phenomena as diverse as the hierarchical assembly of spider silk proteins and the compositional patterns in music. By making these deep similarities explicit, Isomorphic Attention enables AI to reason more like humans do—seeing past surface differences to grasp fundamental patterns that unite seemingly disparate fields. Through this lens, AI systems can learn and generalize, moving beyond superficial pattern matching to true structural understanding. The implications span from scientific discovery to engineering design, offering a new approach to artificial intelligence that mirrors how humans grasp the underlying unity of natural phenomena. Some key insights include: 1️⃣ Graph Isomorphism Neural Networks (GINs): GIN-style aggregation ensures structurally distinct graphs map to distinct embeddings, improving generalization and avoiding relational pattern collapse. 2️⃣ Category Theory Perspective: Transformers as functors preserve structural relationships. Sparse-GIN refines attention into sparse adjacency matrices, unifying domain knowledge across tasks. 3️⃣ Information Bottleneck & Sparsification: Sparsity reduces overfitting by filtering irrelevant edges, aligning with natural systems. Sparse-GIN outperforms dense attention by focusing on crucial connections. 4️⃣ Hierarchical Representation Learning: GIN-Attention captures multiscale patterns, mirroring structures like spider silk. Nested GINs model local and global dependencies across fields. 5️⃣ Practical Impact: Sparse-GIN enables domain-specific fine-tuning atop pre-trained Transformer foundation models, reducing the need for full retraining. Other impacts: ✅Real-World Relevance: Whether we are looking at protein structures, designing new materials, or working on social network analytics, graph-aware Transformers can capture subtle relational patterns traditional architectures may miss. ✅The juncture of graph isomorphism theory, category theory, and sparsification, these GIN-Transformers step beyond sequential modeling to tackle the relational nature of complex data. #Transformers #GraphNeuralNetworks #AI #MachineLearning #Isomorphism #CategoryTheory #ArtificialIntelligence #DeepLearning Link to paper & code in response ⤵️

Markus J. Buehler

19,416 次观看 • 1 年前

[RLHF] by Hand ✍️ Yesterday, Jan Leike (Jan Leike) announced he is joining #Anthropic to lead their "super-alignment" mission. He is the co-inventor of Reinforcement Learning with Human Feedback (#RLHF). How does RLHF work? [1] Given ↳ Reward Model (RM) ↳ Large Language Model (LLM) ↳ Two (Prompt, Next) Pairs 🟪 TRAIN RM Goal: Learn to give higher rewards to winners [2] Preferences ↳ A human reviews the two pairs and picks a "winner" ↳ (doc is, him) Embeddings ↳ This prompt has never received human feedback directly ↳ [S] is the special start symbol [11] Transformer ↳ Attention (yellow) ↳ Feed Forward (4x2 weight and bias matrix) ↳ Output: 3 "transformed" feature vector, one per position ↳ More details in my previous post 8. Transformer [] [12] Output Probabilities ↳ Apply a linear layer to map each transformed feature vector to a probability distribution over the vocabulary. [13] Sample ↳ Apply the greedy method, which is to pick the word with the highest score ↳ For output 1 and 2, the model accurately predicts the next word ↳ For 3rd output position, the model's predicts "him" [14] Reward Model ↳ The new pair (CEO is, him) is fed to the reward model ↳ The process is same as [3]-[6] ↳ Output: Reward = 3 [15] Loss Gradient ↳ We set the loss as the negative of the reward. ↳ The loss gradient is simply a constant -1. ↳ Run backpropagation and gradient descent to update LLM's weights and biases (red border)

[RLHF] by Hand ✍️ Yesterday, Jan Leike (Jan Leike) announced he is joining #Anthropic to lead their "super-alignment" mission. He is the co-inventor of Reinforcement Learning with Human Feedback (#RLHF). How does RLHF work? [1] Given ↳ Reward Model (RM) ↳ Large Language Model (LLM) ↳ Two (Prompt, Next) Pairs 🟪 TRAIN RM Goal: Learn to give higher rewards to winners [2] Preferences ↳ A human reviews the two pairs and picks a "winner" ↳ (doc is, him) Embeddings ↳ This prompt has never received human feedback directly ↳ [S] is the special start symbol [11] Transformer ↳ Attention (yellow) ↳ Feed Forward (4x2 weight and bias matrix) ↳ Output: 3 "transformed" feature vector, one per position ↳ More details in my previous post 8. Transformer [] [12] Output Probabilities ↳ Apply a linear layer to map each transformed feature vector to a probability distribution over the vocabulary. [13] Sample ↳ Apply the greedy method, which is to pick the word with the highest score ↳ For output 1 and 2, the model accurately predicts the next word ↳ For 3rd output position, the model's predicts "him" [14] Reward Model ↳ The new pair (CEO is, him) is fed to the reward model ↳ The process is same as [3]-[6] ↳ Output: Reward = 3 [15] Loss Gradient ↳ We set the loss as the negative of the reward. ↳ The loss gradient is simply a constant -1. ↳ Run backpropagation and gradient descent to update LLM's weights and biases (red border)

Tom Yeh

79,758 次观看 • 2 年前

Imagine 10,000 warehouses, factories, and campuses across India, each sitting on a battery. 🇮🇳 During the day, they store cheap solar power from the grid at Rs 2–3/kWh. At night, they run on it. No diesel. No paying Rs 12/kWh to the DISCOM. And together, those 10,000 batteries can eventually act as multiple clusters of power plants, absorbing surplus when the grid is flooded, releasing it when demand peaks. A virtual power plant, distributed across the country, owned by no one and operated by software. The unit economics work today. The regulation is moving. The solar surplus is already here. Cell prices have fallen 95% in 15 years. A battery at an industrial site today pays for itself in 3–4 years, purely on energy savings. VoltSeal is building the intelligence layer that enables this. Rainmatter by Zerodha is backing them. A short interaction with founder Mudit below 👇

Imagine 10,000 warehouses, factories, and campuses across India, each sitting on a battery. 🇮🇳 During the day, they store cheap solar power from the grid at Rs 2–3/kWh. At night, they run on it. No diesel. No paying Rs 12/kWh to the DISCOM. And together, those 10,000 batteries can eventually act as multiple clusters of power plants, absorbing surplus when the grid is flooded, releasing it when demand peaks. A virtual power plant, distributed across the country, owned by no one and operated by software. The unit economics work today. The regulation is moving. The solar surplus is already here. Cell prices have fallen 95% in 15 years. A battery at an industrial site today pays for itself in 3–4 years, purely on energy savings. VoltSeal is building the intelligence layer that enables this. Rainmatter by Zerodha is backing them. A short interaction with founder Mudit below 👇

Dinesh Pai

15,600 次观看 • 1 个月前

Bio inspired Hebbian probabilistic network learns in less than 5 minutes from a super sparse single reward per episode! also has imitation learning (manual control) system has 3 parallel competing networks which get sensory input from a 360 vision (27-direction sensory neuron array) link to code in comment each sub-network is responsible for a single motor action: forward, left and right. at each step whichever section has most neurons firing wins neurons fire probabilistically and mark themselves with a time-decay tag which happens when a neuron fires and diminishes with time. you can see this " tag countdown" on each neuron when a reward is attained(eating the cheese) eligible connections gets strengthened I included 2 runs in the video first was 15 minutes in real time and second was 5 minutes. red plot is the rolling average of last 10 time to cheese. it is really not possible for agent to achieve full control due to probabilistic neural firing. that is why it has to learn while jittering all over the place, which in itself is interesting in manual mode you can guide the cheese by stimulating its motor control networks ( still probabilistically ) and the rewards will still work ✅ Biologically Plausible Features: Stochastic firing (neurons in the brain fire probabilistically) Reward-based learning (dopamine-like neuromodulation) Hebbian plasticity (well-established biological mechanism) Eligibility traces (biological neurons have temporal credit assignment) Sparse sensory encoding (similar to place cells, grid cells) Competitive action selection (basal ganglia architecture) No backpropagation (which is biologically implausible) ❌ Missing Biological Features: No recurrent connections (real brains have extensive feedback loops) No inhibitory neurons (GABAergic neurons are ~20% of cortex) No spike timing (simplified from true spiking dynamics) Uniform layer structure (biological networks are more heterogeneous) Simple weight updates (real synaptic plasticity is more complex)

Bio inspired Hebbian probabilistic network learns in less than 5 minutes from a super sparse single reward per episode! also has imitation learning (manual control) system has 3 parallel competing networks which get sensory input from a 360 vision (27-direction sensory neuron array) link to code in comment each sub-network is responsible for a single motor action: forward, left and right. at each step whichever section has most neurons firing wins neurons fire probabilistically and mark themselves with a time-decay tag which happens when a neuron fires and diminishes with time. you can see this " tag countdown" on each neuron when a reward is attained(eating the cheese) eligible connections gets strengthened I included 2 runs in the video first was 15 minutes in real time and second was 5 minutes. red plot is the rolling average of last 10 time to cheese. it is really not possible for agent to achieve full control due to probabilistic neural firing. that is why it has to learn while jittering all over the place, which in itself is interesting in manual mode you can guide the cheese by stimulating its motor control networks ( still probabilistically ) and the rewards will still work ✅ Biologically Plausible Features: Stochastic firing (neurons in the brain fire probabilistically) Reward-based learning (dopamine-like neuromodulation) Hebbian plasticity (well-established biological mechanism) Eligibility traces (biological neurons have temporal credit assignment) Sparse sensory encoding (similar to place cells, grid cells) Competitive action selection (basal ganglia architecture) No backpropagation (which is biologically implausible) ❌ Missing Biological Features: No recurrent connections (real brains have extensive feedback loops) No inhibitory neurons (GABAergic neurons are ~20% of cortex) No spike timing (simplified from true spiking dynamics) Uniform layer structure (biological networks are more heterogeneous) Simple weight updates (real synaptic plasticity is more complex)

echo.hive

33,638 次观看 • 8 个月前

This can be the superpower of the Neural Band that Meta is giving together with the Ray-Ban Meta Display glasses. The video shows an old prototype bracelet by CTRL+LABS, the startup acquired by Meta and whose technology was used to develop the Neural Band. At the beginning of the video, the guy makes an action (a keyboard key press) with his hands, then the bracelet can substitute the key pressure, and at the end of the video, the guy doesn't even have to do the action; it is just sufficient that he "thinks" about it. As long as the brain is sending an electric message to the fingers, the full action is not necessary anymore. Just an "intention" to move them is necessary. If the Neural Band is evolved to this stage, and the users are educated to this, potentially, we may not even need to perform air taps or writing gestures, but we could just think about doing them. This would reduce a lot of the fatigue of using XR devices and the weirdness of using them on the street. Then why isn't this feature available today? I guess that the reason is twofold. First of all, we have accuracy: the full gesture is easier to detect for the system. Many people (me included) are praising the accuracy of the Neural Band, and this is amazing, because an input mechanism should have a reliability close to 100%. Then we, as users, have never been trained to just "think" about actions: it would feel weird and hard to learn. I think we should undergo some training to learn how to do this "thinking" operation properly. I hope that something like this could come in the upcoming years... that would be the real game-changer paradigm if compared to the camera-based tracking.

This can be the superpower of the Neural Band that Meta is giving together with the Ray-Ban Meta Display glasses. The video shows an old prototype bracelet by CTRL+LABS, the startup acquired by Meta and whose technology was used to develop the Neural Band. At the beginning of the video, the guy makes an action (a keyboard key press) with his hands, then the bracelet can substitute the key pressure, and at the end of the video, the guy doesn't even have to do the action; it is just sufficient that he "thinks" about it. As long as the brain is sending an electric message to the fingers, the full action is not necessary anymore. Just an "intention" to move them is necessary. If the Neural Band is evolved to this stage, and the users are educated to this, potentially, we may not even need to perform air taps or writing gestures, but we could just think about doing them. This would reduce a lot of the fatigue of using XR devices and the weirdness of using them on the street. Then why isn't this feature available today? I guess that the reason is twofold. First of all, we have accuracy: the full gesture is easier to detect for the system. Many people (me included) are praising the accuracy of the Neural Band, and this is amazing, because an input mechanism should have a reliability close to 100%. Then we, as users, have never been trained to just "think" about actions: it would feel weird and hard to learn. I think we should undergo some training to learn how to do this "thinking" operation properly. I hope that something like this could come in the upcoming years... that would be the real game-changer paradigm if compared to the camera-based tracking.

TonyVT SkarredGhost

11,805 次观看 • 9 个月前

The Mathematics of Moving a Cursor with Neural Signals What might Neuralink Neuralink be doing Mathematically? Consider the task of moving a cursor without touching it. The machine is not looking for a full thought, a sentence, or an image. For this Control problem, the useful object is an intended movement state. sₜ = (pₜ, vₜ) Here, pₜ is the cursor position at time t, and vₜ is the velocity the user is trying to express. The implant records neural activity through many electrode channels, then the decoder tries to estimate vₜ from that activity. Neuralink’s PRIME material describes the N1 Implant as recording and transmitting brain activity with the goal of enabling computer control. For channel i, a simple population model is rᵢ(t) ≈ bᵢ + aᵢ max(0, dᵢ · vₜ) + ηᵢ(t) where rᵢ(t) is the measured activity, bᵢ is baseline activity, aᵢ is channel gain, dᵢ is the channel’s preferred movement direction, and ηᵢ(t) is noise. One channel is not the command. The useful signal is the pattern across many channels: rₜ = (r₁(t), r₂(t), …, rₙ(t)) The decoder subtracts the baseline vector b and applies a learned map W: v̂ₜ = W(rₜ − b) This gives an estimate of the intended velocity. The cursor then updates by pₜ₊₁ = pₜ + Δt v̂ₜ This is the loop shown in the render: neural activity -> decoded velocity -> cursor motion The cortical network and electrode threads show the measurement side. The N1 Implant is described as using 1,024 electrodes distributed across 64 flexible threads, each thinner than a human hair. The decoder panel shows the computational side with activity rₜ, decoded velocity v̂ₜ, and the cursor state pₜ changing over time. A noisy biological pattern becomes a state estimate. That estimate becomes motion on a screen. Therefore, the first lesson is not that Neuralink makes the brain a screen. For cursor control, the Mathematics is more precise: A small piece of intention is represented as a hidden state, measured through neural activity, decoded as a vector, and turned into action. #Neuralink #BrainComputerInterface #NeuralEngineering #Mathematics #StateEstimation #Neuroscience #MachineLearning #BiomedicalEngineering

The Mathematics of Moving a Cursor with Neural Signals What might Neuralink Neuralink be doing Mathematically? Consider the task of moving a cursor without touching it. The machine is not looking for a full thought, a sentence, or an image. For this Control problem, the useful object is an intended movement state. sₜ = (pₜ, vₜ) Here, pₜ is the cursor position at time t, and vₜ is the velocity the user is trying to express. The implant records neural activity through many electrode channels, then the decoder tries to estimate vₜ from that activity. Neuralink’s PRIME material describes the N1 Implant as recording and transmitting brain activity with the goal of enabling computer control. For channel i, a simple population model is rᵢ(t) ≈ bᵢ + aᵢ max(0, dᵢ · vₜ) + ηᵢ(t) where rᵢ(t) is the measured activity, bᵢ is baseline activity, aᵢ is channel gain, dᵢ is the channel’s preferred movement direction, and ηᵢ(t) is noise. One channel is not the command. The useful signal is the pattern across many channels: rₜ = (r₁(t), r₂(t), …, rₙ(t)) The decoder subtracts the baseline vector b and applies a learned map W: v̂ₜ = W(rₜ − b) This gives an estimate of the intended velocity. The cursor then updates by pₜ₊₁ = pₜ + Δt v̂ₜ This is the loop shown in the render: neural activity -> decoded velocity -> cursor motion The cortical network and electrode threads show the measurement side. The N1 Implant is described as using 1,024 electrodes distributed across 64 flexible threads, each thinner than a human hair. The decoder panel shows the computational side with activity rₜ, decoded velocity v̂ₜ, and the cursor state pₜ changing over time. A noisy biological pattern becomes a state estimate. That estimate becomes motion on a screen. Therefore, the first lesson is not that Neuralink makes the brain a screen. For cursor control, the Mathematics is more precise: A small piece of intention is represented as a hidden state, measured through neural activity, decoded as a vector, and turned into action. #Neuralink #BrainComputerInterface #NeuralEngineering #Mathematics #StateEstimation #Neuroscience #MachineLearning #BiomedicalEngineering

Mathelirium

14,424 次观看 • 2 个月前

Andrew Huberman drops raw truth on self-mastery: The biggest mistake most people make is hunting for the external fix—caffeine, activities, new routines—while the real power to change everything lives inside. “You realize the ability to withhold reflexes you don’t want—like getting your temper sparked, not speaking from your first thought but your second or third—all of that comes from internal work.” He admits he spent years chasing outside solutions before seeing the pattern: Thoughts layer like embers in a fire—you can feed them or starve them. Rumination on something painful? Distraction is often the smarter move unless you’re deliberately processing it. The brain is wild with input—stopping, listening, and going “wow, my mind is crazy” is where the shift starts. Clip from this 1:35 clip—pure neuroscience-meets-wisdom gold. The most powerful upgrade isn’t something you add... it’s something you learn to not do. What’s one reflex or thought pattern you’ve started catching and starving instead of feeding? Your wins (or ongoing battles)—drop them below.

Andrew Huberman drops raw truth on self-mastery: The biggest mistake most people make is hunting for the external fix—caffeine, activities, new routines—while the real power to change everything lives inside. “You realize the ability to withhold reflexes you don’t want—like getting your temper sparked, not speaking from your first thought but your second or third—all of that comes from internal work.” He admits he spent years chasing outside solutions before seeing the pattern: Thoughts layer like embers in a fire—you can feed them or starve them. Rumination on something painful? Distraction is often the smarter move unless you’re deliberately processing it. The brain is wild with input—stopping, listening, and going “wow, my mind is crazy” is where the shift starts. Clip from this 1:35 clip—pure neuroscience-meets-wisdom gold. The most powerful upgrade isn’t something you add... it’s something you learn to not do. What’s one reflex or thought pattern you’ve started catching and starving instead of feeding? Your wins (or ongoing battles)—drop them below.

Camus

88,621 次观看 • 4 个月前

I analyzed 107 market crashes from 1929 to 2025 and found that 94% follow one of four recovery patterns. Before sharing them with you, let's understand why it's important to learn about them. If you can predict the type of market recovery pattern, you can predict movements and timelines. If you can predict the timeline, you can predict exactly when to enter and exit a trade. And ultimately, you can profit from any market crash. Here are the 4 recovery patterns I’ve identified: Pattern #1: V-Shaped Recovery (18% of crashes) The market crashes hard and bounces back fast. Everyone hopes for this, but it's rare—less than one in five crashes. Pattern #2: U-Shaped Recovery (Most Common) The market crashes, stays down for 6-18 months, then recovers. This is what actually happens most of the time. Pattern #3: W-Shaped Recovery (The Double Dip) The market crashes, starts recovering (fake out), then crashes harder. This one destroys people emotionally because they get hit twice—buying the first dip, then selling the second crash. Pattern #4: L-Shaped Recovery (7% of crashes) The market crashes and stays depressed for years. Japan 1989 took 30 years to recover. But this is extremely rare. — This is just one of the topics covered from my 19-minute recording on analyzing market crashes. We also talked about the optimal buying window (when to enter for maximum returns), which sectors recover first, and how to scale in without going all-in too early. Just comment "CRASH" and I'll send you the full video if you'd like to watch it.

I analyzed 107 market crashes from 1929 to 2025 and found that 94% follow one of four recovery patterns. Before sharing them with you, let's understand why it's important to learn about them. If you can predict the type of market recovery pattern, you can predict movements and timelines. If you can predict the timeline, you can predict exactly when to enter and exit a trade. And ultimately, you can profit from any market crash. Here are the 4 recovery patterns I’ve identified: Pattern #1: V-Shaped Recovery (18% of crashes) The market crashes hard and bounces back fast. Everyone hopes for this, but it's rare—less than one in five crashes. Pattern #2: U-Shaped Recovery (Most Common) The market crashes, stays down for 6-18 months, then recovers. This is what actually happens most of the time. Pattern #3: W-Shaped Recovery (The Double Dip) The market crashes, starts recovering (fake out), then crashes harder. This one destroys people emotionally because they get hit twice—buying the first dip, then selling the second crash. Pattern #4: L-Shaped Recovery (7% of crashes) The market crashes and stays depressed for years. Japan 1989 took 30 years to recover. But this is extremely rare. — This is just one of the topics covered from my 19-minute recording on analyzing market crashes. We also talked about the optimal buying window (when to enter for maximum returns), which sectors recover first, and how to scale in without going all-in too early. Just comment "CRASH" and I'll send you the full video if you'd like to watch it.

Felix Prehn 🐶

17,194 次观看 • 7 个月前

A new Nature paper from Johns Hopkins (by Prof. Lin Dingchang Lin ) just solved one of the hardest problems in biology: how do you record what every cell in a tissue experienced over time, not just what it looks like right now? The answer: GEMINI — Granularly Expanding Memory for Intracellular Narrative Integration. It works exactly like tree rings. Cells are genetically engineered to express a computationally designed protein assembly. As the assembly grows inside the cell, it captures cellular activity as fluorescent ring patterns — each ring a timestamp, each ring's properties encoding signal intensity. Look at a cross-section under a microscope and you can read the cell's history backward, with ~15-minute resolution. The key: cells build the recorder themselves. GEMINI doesn't interfere with normal function — it just quietly writes. What they demonstrated: In a full tumor xenograft, GEMINI captured every cancer cell's activity history across the entire tumor while it continued to grow normally. For the first time, researchers can look back and see how different regions of the same tumor responded differently to therapy over time — not snapshots, but film. In a mouse brain, GEMINI recorded neural activity dynamics without disrupting behavior, coordination, or memory. It could temporally resolve the history of a brain seizure. Why this matters: Every tool we have in biology gives you state — what the cell looks like now. Sequencing, imaging, proteomics — all snapshots. GEMINI gives you trajectory. It's the difference between a photograph and a video, applied to every cell in an organ simultaneously. The team is explicit that AI-based decoding tools will be central to reading GEMINI's output at whole-brain scale. This is the data layer that makes temporal single-cell atlases possible. Paper: Congratulations Dingchang Lin

A new Nature paper from Johns Hopkins (by Prof. Lin Dingchang Lin ) just solved one of the hardest problems in biology: how do you record what every cell in a tissue experienced over time, not just what it looks like right now? The answer: GEMINI — Granularly Expanding Memory for Intracellular Narrative Integration. It works exactly like tree rings. Cells are genetically engineered to express a computationally designed protein assembly. As the assembly grows inside the cell, it captures cellular activity as fluorescent ring patterns — each ring a timestamp, each ring's properties encoding signal intensity. Look at a cross-section under a microscope and you can read the cell's history backward, with ~15-minute resolution. The key: cells build the recorder themselves. GEMINI doesn't interfere with normal function — it just quietly writes. What they demonstrated: In a full tumor xenograft, GEMINI captured every cancer cell's activity history across the entire tumor while it continued to grow normally. For the first time, researchers can look back and see how different regions of the same tumor responded differently to therapy over time — not snapshots, but film. In a mouse brain, GEMINI recorded neural activity dynamics without disrupting behavior, coordination, or memory. It could temporally resolve the history of a brain seizure. Why this matters: Every tool we have in biology gives you state — what the cell looks like now. Sequencing, imaging, proteomics — all snapshots. GEMINI gives you trajectory. It's the difference between a photograph and a video, applied to every cell in an organ simultaneously. The team is explicit that AI-based decoding tools will be central to reading GEMINI's output at whole-brain scale. This is the data layer that makes temporal single-cell atlases possible. Paper: Congratulations Dingchang Lin

Bo Wang

85,108 次观看 • 4 个月前

Always so much fun to chat with Grant Sanderson AI has been making much faster progress in math than in other fields. As a result, mathematics is showing us, very concretely, what AI progress in other fields will look like. Even within mathematics, there's a jagged landscape. What does it look like? What is the nature of the most important conceptual breakthroughs in the history of mathematics, and how different are they from what AIs are currently able to do? Does AI (on net) increase or decrease human understanding of the field? How big is the overhang from having AIs systematically try to connect ideas already in the literature? And what advice does Grant have for aspiring mathematicians, coders, and other students who are passionate about fields that are being most transformed upon by AI? 0:00:00 – AI is discovering new proofs. Is that AGI? 0:11:32 – The verification loop on conceptual breakthroughs can be a century long 0:26:12 – Will we understand an AI proof of the Riemann hypothesis? 0:38:08 – Can AI find the hidden bridges between fields? 0:53:48 – Why real-world tasks don’t fit into RL environments 1:07:07 – Good writing requires theory of mind that AI still lacks 1:16:02 – Why learning will still depend on human curation Look up Dwarkesh Podcast on Spotify, Apple Podcasts, YouTube, etc.

Always so much fun to chat with Grant Sanderson AI has been making much faster progress in math than in other fields. As a result, mathematics is showing us, very concretely, what AI progress in other fields will look like. Even within mathematics, there's a jagged landscape. What does it look like? What is the nature of the most important conceptual breakthroughs in the history of mathematics, and how different are they from what AIs are currently able to do? Does AI (on net) increase or decrease human understanding of the field? How big is the overhang from having AIs systematically try to connect ideas already in the literature? And what advice does Grant have for aspiring mathematicians, coders, and other students who are passionate about fields that are being most transformed upon by AI? 0:00:00 – AI is discovering new proofs. Is that AGI? 0:11:32 – The verification loop on conceptual breakthroughs can be a century long 0:26:12 – Will we understand an AI proof of the Riemann hypothesis? 0:38:08 – Can AI find the hidden bridges between fields? 0:53:48 – Why real-world tasks don’t fit into RL environments 1:07:07 – Good writing requires theory of mind that AI still lacks 1:16:02 – Why learning will still depend on human curation Look up Dwarkesh Podcast on Spotify, Apple Podcasts, YouTube, etc.

Dwarkesh Patel

288,448 次观看 • 25 天前

Did you know that in Blender you can have edge highlighting and ambient occlusion right in the regular 'Solid' shading mode? ✨ People were asking about my viewport shading settings, so here they are! 1. Press the small button on the top right of your viewport 2. Check 'Cavity', and set the type to 'Both'. World Space 'Valley' acts exactly like ambient occlusion. I like to disable the 'Ridge' here though, because inverted AO doesn't make sense to me visually 3. I also usually keep the 'Shadow' enabled for extra definition, but keep in mind that this is performance-heavy, and if you have millions of polys you may see stuttering 4. I also like to enable 'Backface Culling'. It's not uncommon to produce inverted geometry when you model. This setting allows me to see it immediately, and thanks to that, my geometry is always correct when I export it to Unity! 5. You can also select a different studio lighting preset, or go with a matcap if that's something you're into! 🚀 #b3d

Did you know that in Blender you can have edge highlighting and ambient occlusion right in the regular 'Solid' shading mode? ✨ People were asking about my viewport shading settings, so here they are! 1. Press the small button on the top right of your viewport 2. Check 'Cavity', and set the type to 'Both'. World Space 'Valley' acts exactly like ambient occlusion. I like to disable the 'Ridge' here though, because inverted AO doesn't make sense to me visually 3. I also usually keep the 'Shadow' enabled for extra definition, but keep in mind that this is performance-heavy, and if you have millions of polys you may see stuttering 4. I also like to enable 'Backface Culling'. It's not uncommon to produce inverted geometry when you model. This setting allows me to see it immediately, and thanks to that, my geometry is always correct when I export it to Unity! 5. You can also select a different studio lighting preset, or go with a matcap if that's something you're into! 🚀 #b3d

passivestar

90,380 次观看 • 3 年前