Congratulations Giannis Daras! I worked a lot on diffusion + corrupted measurements in my PhD. During that time, it has been a pleasure seeing your work tackle similar problems from different angles, and taking the topic to new heights!

Accepted to #ICML2024!! 🚀 Meet MAIA- a Multimodal Automated Interpretabilty Agent that helps users understand AI systems. Given a user query (eg "label a model’s feature"), MAIA designs experiments iteratively by forming and updating hypotheses based on experimental results.

Congratulations on the ICLR best paper award! Very well deserved. This paper sheds light on under-explored aspects of diffusion models, and provides important observations. Amazing work by Zahra Kadkhodaie Florentin Guth Eero Simoncelli Stephane Mallat

This is cool.. new Vision OS 2 will be able to transform your photos into 3D! #wwdc

Was super fun to work on this, with many talented people! Excited to see the reactions to old photos being viewed spatially 😊

🤯 legit mind blown moment. Trying out the new #AppleVisionPro beta that lets you turn almost any photo into a 3D "spatial photo". Impossible to convey over a phone screen, but seeing moments from the past like this is just incredibly immersive / emotional. Pretty darn cool.

I am absolutely blown away by the new visionOS 2 feature that turns regular photos into spatial photos. It works even better than taking actual spatial photos and video using my iPhone 15 Pro Max. I expected wonky and inconsistent cutting out of subjects. There is none. 10/10!

Extremely proud to be working on this with many amazing people Luma AI! Generate a 5-second, 120 frames video in 120 seconds from text or images *now* on: lumalabs.ai/dream-machine Available to everyone. #LumaDreamMachine

To be fair, the exact token count can be slightly different, as the tokener design in videos has not fully matured. But it is safe to say that the size (on disk) of the video datasets are hundreds times bigger than text datasets -- scaling can work quite differently here!

I will be attending #ICML2024 in Vienna. Happy to talk and catch up with old and new friends 😊 DM me if you want to meet!

#ICML2024 diffusion circle 📢: let's meet at 3PM on Thursday at registration, near the Vienna tourism desk. We'll find a spot to sit nearby. Spread the word, bring your diffusion friends!

Second down: D̴i̴f̴f̴u̴s̴i̴o̴n̴ ̴x̴ ̴M̴u̴l̴t̴i̴m̴o̴d̴a̴l̴ Great seeing many familiar faces! Stefano Ermon Ben Poole Jiajun Wu Soumith Chintala Sander Dieleman Jessy Lin vicc Robin Rombach Tanishq Mathew Abraham, Ph.D. Karsten Kreis and more!! And big congrats Dan Kondratyuk on winning best

Tomorrow (Saturday), I'm giving a talk about the intuition behind diffusion models for audiovisual data at the #ICML2024 controllable video generation workshop, at 2:30PM! I'm aiming for it to be accessible to everyone, but hopefully also useful for seasoned practitioners.

Ph. Done! 🎓🥳 While it is a year late, it's nice to celebrate and reflect on this journey, which started in 2020. 3.5 years and 12 papers later, I learned a lot, made many friends, met some amazing minds, and hopefully, made some impact on fellow and future researchers.