Lexunit’s Post

Generative Models: What do they know? Do they know things? Amazing new paper that addresses the questions we are all heavily debating since the Sora release: When Sora was released I was one of the enthusiast talking about how Sora has to implicitly know geometric properties and physical behavior of the objects in the scene. In contrast, many critics pointed out the flaws in the displayed physics. Most notably, Yann LeCun very convincingly argued that "modeling the world for action by generating pixel is [...] wasteful and doomed to failure". My expertise is definitely not deep enough to contradict that, but the topic absolutely fascinates me. Now I came across this very interesting paper: "Generative Models: What do they know? Do they know things? Let's find out!" by Xiaodan Du et al. : https://lnkd.in/evq5bpCf In the paper, the authors show that generative models indeed learn some intrinsic features of a scene such as normals or depth maps 😲. #sora #physics #ai

Diffusion Models: Midjourney, Dall-E Reverse Time to Generate Images from Prompts: We have AI, physics, thermodynamics, probability, beautiful images AND time travel? What are you waiting for. Continue reading on Towards Data Science » #MachineLearning #ArtificialIntelligence #DataScience

The realism in some of the OpenAI Sora videos is unparalleled. It also brings up an interesting problem of object physics and interaction as can be seen in this video. Might be interesting to try converting patches into an object with physical attributes and compute its simulation using the Chaos system in Unreal Engine. This might help generate particle interactions and material motion better. OpenAI Prompt: Archeologists discover a generic plastic chair in the desert, excavating and dusting it with great care. #AI

#Technology #DataAnalytics #DataDriven Diffusion Models: Midjourney, Dall-E Reverse Time to Generate Images from Prompts: We have AI, physics, thermodynamics, probability, beautiful images AND time travel? What are you waiting for. Continue reading on Towards Data Science » #MachineLearning #ArtificialIntelligence #DataScience

🌌 Exciting developments in the intersection of AI and physics! 🚀 Discover how artificial intelligence is revolutionizing the field, from enhancing research efficiencies to deepening our understanding of the universe. 🔍 AI is transforming everything from high-energy physics to quantum mechanics, opening new frontiers and accelerating scientific discovery. Whether you're a scientist, engineer, or tech enthusiast, this article is a must-read! 👉 Let's explore the future possibilities that AI brings to physics! #AI #Physics #Innovation #ScienceResearch

https://lnkd.in/gs3VV7Sv Theory of relativity have primarily been validated through classical experimental methods, and Artificial Intelligence (AI) models are not typically used to propose or validate these theories. However, AI and computational methods can play important roles in related areas. AI techniques may assist in analyzing large datasets from astronomical observations. AI is employed in the analysis of data from gravitational wave detectors, such as those used by LIGO and Virgo. These detectors directly observe phenomena predicted by the General Theory of Relativity, such as the merging of black holes or neutron stars. AI models can be used to simulate complex cosmological scenarios based on the principles of relativity. These simulations help researchers understand the large-scale structure of the universe and the behavior of matter and energy on cosmic scales. #machinelearning #deeplearning #spacescience #relativity #einstein #artificialintelligence

Google’s new weather prediction system combines AI with traditional physics. While new machine-learning techniques are extremely fast and efficient, they struggle with long-term predictions. General circulation models, which have dominated weather prediction for the last 50 years, use complex PDEs and give more accurate longish term projections, but they are exceedingly slow and expensive to run. The new model, called NeuralGCM, attempts to combine the two.

https://lnkd.in/g4Fn2Z7Z Very interesting interview. Small snippet: ideas borrowed from physics help advance AI research

OpenAI Sora isn’t just a creative plaything compared to DALLE, Midjourney, Leonardo or Stable Diffusion; it’s a leap further. Sora is a data-fueled physics engine that can do: 1. Text to video 2. Image to video 3. Video to video 4. Up to 1 minute videos (one single one or extend) Sora brings to life diverse universes, whether real or fantastical. It excels in sophisticated rendering, understanding of physics, and ensuring consistency over time, all through cutting-edge denoising methods and gradient mathematics. #openai #ai #artificialintelligence #disruption