Real-Time Radiance Field Rendering (huggingface.co)
Achieves SOTA on quality AND on training time AND renders in real-time (60fps+)
Divide & Bind technique for generative image models (sites.google.com)
Greatly improves Stable Diffusion's issues of missing objects and mixing up attributes
CALM: Conditional Adversarial Latent Models for Directable Virtual Characters (research.nvidia.com)
Siggraph 2023, Nvidia improves on their previous research into controllable, natural movement learnt from unlabelled data. Code and paper available.
OC When you ask someone to check your code
OC The midpoint of the debugging journey
Interview with Inflection AI co-founder and CEO Mustafa Suleyman (www.barrons.com)
"We are about to train models that are 10 times larger than the cutting edge GPT-4 and then 100 times larger than GPT-4. That’s what things look like over the next 18 months."
RT-2: DeepMind robotic research based on PaLM visual language models (www.deepmind.com)
Up to 100% improvement on unseen tasks, environments, and backgrounds
OC Looking at the code for a legacy project
OC The scope creep is coming from inside the house
Universal and Transferable Attacks on Aligned Language Models (llm-attacks.org)
OC Die with dignity, for goodness sake
OC C, can we have closures?
What Frame Rate is Needed to Simulate Reality? (www.youtube.com)
The frame rate needed to completely eliminate all the artifacts associated with discrete frame rates
Tunable Stiffness Compliant Mechanism with Bistable Switch (www.youtube.com)
Translational binary stiffness compliant mechanism that achieves two different states of stiffness by being triggered using a simple bistable switch
Lego-like Compliant-mechanism Building Blocks that Maintain their DOFs (www.youtube.com)
Configuration-indifferent compliant building blocks that can be assembled like Lego in any configuration or arrangement to produce compliant mechanisms of any complexity that achieve the same degrees of freedom (DOFs) as their constituent building blocks.
Compliant Mechanisms that learn - Mechanical Neural Network Architected Materials (www.youtube.com)
The world’s first mechanical neural network that can learn its behavior. It consists of a lattice of compliant mechanisms that constitute an artificial intelligent (AI) architected material that gets better and better at acquiring desired behaviors and properties with increased exposure to unanticipated ambient loading...
AnimateDiff: New Approach For Animating Diffusion Models Without Specific Tuning (animatediff.github.io)
OC But if that's not the solution, then what is?
OC It's the only logical explanation
OC Ah, General Kenobi
OC Macros have only ever made my life easier
OC Just finishing up a PR
A tiny RWKV model with 2.9M (!) params can solve 4.2379*564.778-1209.01 with CoT (twitter.com)
Control-A-Video: Controllable Text-to-Video Generation with Diffusion Models (controlavideo.github.io)
Open source "Controlnet for video" compatible with any stable-diffusion-v1-5 based model