🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Netflix just dropped Go-with-the-Flow Motion-Controllable Video Diffusion Models Using Real-Time Warped Noise

Sam Altman: "we have been on the wrong side of history" with regards to open source/open weights AI models

This layer in DINO-v2 dedicates about half its attention mass to a single operation. Each of the sixteen heads independently learns the same circuit to perform this task. What is this all-important operation? The “no-op”. That’s right, we’re spending half our computation to do…

📢📢📢 "𝐑𝐚𝐝𝐢𝐚𝐧𝐭 𝐅𝐨𝐚𝐦: Real-Time Differentiable Ray Tracing", a mesh-based 3D represention. radfoam.github.io arxiv.org/abs/2502.01157 Co-lead by my PhD students Shrisudhan Govindarajan and Daniel Rebain, and w/ Kwang Moo Yi

Hypernetwork Fields was accepted to #CVPR2025 🎉

we trained a new model that is good at creative writing (not sure yet how/when it will get released). this is the first time i have been really struck by something written by AI; it got the vibe of metafiction so right. PROMPT: Please write a metafictional literary short story

📢 "StochasticSplats: Stochastic Rasterization for Sorting-Free 3D Gaussian Splatting" 📢 Project page: github.com/ubc-vision/sto… arxiv.org/abs/2503.24366 Introducing sort-free and pop-free stochastic rendering and training of 3D Gaussian.

The attention layers in the VAEs for FLUX, Stable Diffusion 3.5, and SDXL don't do anything. You can ablate them with almost no effect. At first I thought they might be involved in some clever circuitry—maybe moving global information—but no they're just flailing around doing

Our universe may be the time-reversed interior of a black hole, with the Big Bang as the singularity. The arrow of time follows increasing entropy, which is why we experience time as moving away from the Big Bang. Source: youtube.com/watch?v=A8bBhk…

Group norm is a destructive operation. It normalizes out much of the information regarding the relative magnitudes of channels. But that information is important! In this VAE many of those channels are describing colors—imagine what would happen if you normalized each channel of

The recent launches of Starship remind me of this Calvin and Hobbes comic

this gives a pretty good explanation into how models learn in particular, it explains grokking grokking occurs *exactly* when capacity saturates. this is where models can't perfectly fit every training example, so they have to share info bt examples in a smart way

Last week, our Triangle splatting paper was quietly released, and since then the tech community ignited fierce debates about it ! It was trending on Hacker News ! Today we released the code! A deep dive into the epic “comeback” of Triangles to the throne of 3D 🧵 1/n

Come to poster #103 tomorrow morning to find out how to make hypernetworks stable and scalable #CVPR25 x.com/IAmEricHedlin/…