Slew Shy

@okazz_ particles and sine mathematics shine eyecandy

@84Aleha I've been to this street with the trees and panel buildings...

👤: How detailed is your art? Me: Yes

@Peakanime0 indeed was a great time to be yоungster. all was magic, computers and animation including.

Goodness Gracious … it’s crazy how the closer you get to the 90s, the animation just keeps getting better… like this was the peak of pure hand-drawn craft before everything started changing

@ai_bites looks quadtree compression to me, but gr8t stuff gonna try it.

Adaptive Patch Transformers (APT), a method to accelerate vision transformers (ViTs) by using multiple different patch sizes within the same image. APT reduces the total number of input tokens by using larger patch sizes in more homogeneous image regions, and smaller patches in more complex ones. APT achieves a drastic speedup in ViT inference and training, increasing throughput by 40% on ViT-L and 50% on ViT-H while maintaining downstream performance. It can be applied to a previously fine-tuned ViT and converges in as little as 1 epoch, enabling training on high-resolution images with minimal compute budgets. It also significantly reduces training and inference time with no performance degradation on high-resolution dense visual tasks, achieving up to 30% faster training and inference on visual QA, object detection and semantic segmentation. Paper Title: Accelerating Vision Transformers with Adaptive Patch Sizes Project: rccchoudhury.github.io/apt/ Link: arxiv.org/abs/2510.18091

@Andercot Consciousness is very likely a sub-product, something that manifest through the body, thanks to the body, and not before or after it.

Intelligence is a system level description of emergent coordinated action in response to the environment, and it does not require any 'embodied self'

@VideoGameHstry Thank God it didnt.

If Super Mario Bros. was made in Unreal Engine

@kattlatte neat how diffusion draws a different hair on the second scene. so unconvincing.

ready for the day 👹💅✨

@ShimazuSystems very fast until you hit an overbooked cluster, right in the middle of this giant RAG ingest, and then suddenly tis not.

well it is very, very fast

@tritlo e-screens are the future

e-ink is so good for being able to enjoy the sun without sacrificing your eyes

@stevibe both are amazing if u know how to talk to them. truly marvelous achievement of humanity.

I gave two MoE models the same vibe coding challenge Qwen3.6 35B A3B (31.8GB) vs Gemma4 26B A4B (23.3GB) Stack: > Unsloth Q6_K_XL > llama.cpp > Model-card recommended sampling for each 4 prompts, side-by-side. Which one do you think wins?

@YUDHO_XYZ x.com/SlewShy/status…

Pixels breathe.

@AhmedShahnab :D perhaps for a start have the lat/lon projected on the mesh (probably with certain error), but still good for planetary scale.

@SlewShy Now your question make me wonder as well.. hmmmm

Transparent Globe Project- It’s finally complete! 10 unique interactive 3D globes, each visualizing a completely different global dataset in real time. From river flows and seismic activity to population density, global inequality, satellite constellations, and even Sonic Nomad roaming the planet with live music. Every globe is transparent, responsive, and built from scratch to feel alive. #CreativeCoding @threejs @reactthreefiber #DataVisualization #WebGL #GenerativeArt #BuildInPublic

@MaziyarPanahi @GoogleDeepMind @AIatMeta @lmstudio @Prince_Canuma Kudos to them all, these things actually work. As we speak, and on this laptop this message got written.

SAM 3 generates pixel-accurate masks from text prompts. No manual selection. Gemma 4 decides what to segment and how to style it. SAM 3 executes. Gemma 4 by @GoogleDeepMind SAM by @AIatMeta Local Inference by @lmstudio mlx-vlm engine by @Prince_Canuma

Gemma 4 sees a kid and three dogs. Decides what matters. Calls SAM 3.1 Mask and bounding box. Spotlight on subjects. Background blur. Background pixelation. Four effects. Fully agentic. Two models talking to each other on a MacBook. No App. No cloud. What would you edit?

@EmanueleRodola Trackers rule for life.

formally proved in Lean4 that music is not Turing complete. that's because there can be no endogenous self-replication during music playback (unfortunately). proof: every infinite playback of any valid midi-like symbolic composition is eventually periodic.

@francoisfleuret Kudos! Kudos every time a brave soul dares explain vector fields, manifolds and the n-dim batching visually! THIS is how you teach ML, not stupid squares with arrows and cryptic medieval scribbles.

Reminder that I wrote a little book about deep-learning, which is phone-formatted, entirely free, and nearing the 1M download: fleuret.org/francois/lbdl.…

@yacinelearning would've been nice to show a cloud of points in n-dims and how they travel the manifold

I love it so much when an AI architecture paper is showing said architecture

@msurguy @andreasgysin @lennyjpg also not new, for me it started ~2 years ago

@andreasgysin @lennyjpg Let’s make a website where we spearhead this new art form? I’m also making tiny displays with generative art

Tiny screens for tiny generative art ↓

@msurguy @andreasgysin @lennyjpg lets call it embeddoscene :D

@thefinnmckenty how come are they so clean - no rubber, no scratching?

Before I knew what design was, I was into graffiti. I spent THOUSANDS of hours drawing letters like this old sketchbook I just found from ~2009. Since you have to go really slow (vs designing on a computer), it forces you to really think through all the little details that you might normally skip. How many different ways can you do the descender on a G? Which ways look cool, and why? What can you do with the crossbar of an A when it’s after a K vs a W? Etc etc. The biggest thing you learn is that there are NO SHORTCUTS to mastering anything. It’s going to be hard and take a long ass time. The only way to get really good at drawing Rs is to draw 1000s of Rs…. that’s it.