[1/N] Generative AI has revolutionized how we create text and images. How about designing novel materials? We at @MSFTResearch #AI4Science are thrilled to announce MatterGen: our generative model that enables broad property-guided materials design. 👇 arxiv.org/abs/2312.03687

@RRejeleene yes!

IBM & Meta are launching the AI Alliance to advance *open* & reliable AI. The list of over 50 founding members from industry, government, and academia include AMD, Anyscale, CERN, Hugging Face, the Linux Foundation, NASA.... ai.meta.com/blog/ai-allian…

Work with us at the Geometric Learning Lab @KhouryCollege @Northeastern fall 2024! Recruiting 2 PhD students to study geometric structure in deep learning. Learn more: geometriclearninglab.com Apply by Dec. 15: khoury.northeastern.edu/apply/phd-appl…

@roydanroy @mmbronstein @qberthet @tszzl @Ahmed_AI035 If designed correctly, equivariant nets are actually easier to optimize than non equivariant ones, because every parameter gets a less noisy gradient (as if we had a batch of all augmented inputs). So eg it takes far fewer grad steps to optimize a GCNN than plain CNN. Some… twitter.com/i/web/status/1…

I am asking all prominent academic and industry dog breeders to sign my open letter demanding a six month pause in breeding given the clear and present existential risks. twitter.com/lorenlugosch/s…

Is anything known about the characteristics of a data distribution that are required to get scaling laws? Are there any examples of distributions & expressive model classes where one does not get predictable scaling?

Code released for Geometric Algebra Transformers! twitter.com/johannbrehmer/…

Props to this team for running experiments regarding AI safety instead of merely arguing in the abstract. But the conclusions are *way* overstated IMHO. They show that finetuning a model on a virology dataset lets it answer questions about virology, and that this holds even if… twitter.com/i/web/status/1…

Another stunning blogpost by @maurice_weiler to explain his new book (which you should read if you want the ultimate intro to equivariant DL). twitter.com/maurice_weiler…

@mentzer_f What’s the Kolmogorov complexity of this one?

@_onionesque Well said

An interesting aspect of this discussion is the fact that LLMs will soon start affecting our thoughts, beliefs, mental & linguistic habits, and culture. The idea that we could select a handful of "trustworthy" institutions with the "correct" set of values and beliefs to shape LLM… twitter.com/i/web/status/1…

"Get ready to drop llama-3. It's time we finish them"

@AravSrinivas The OG paperclip maximizer

@amaldorai It’s about compute. No amount of money can get you an world class cluster on day 1

We have heard many extrapolations of Mistral AI’s position on the AI Act, so I’ll clarify. In its early form, the AI Act was a text about product safety. Product safety laws are beneficial to consumers. Poorly designed use of automated decision-making systems can cause… twitter.com/i/web/status/1…

In case you don't recognise the name, Vijay was Stanford Professor of Chemistry *and* Structural Biology *and* Computer Science prior to joining A16Z. Few other people in the world have such a relevant background for this particular question IMO. twitter.com/vijaypande/sta…

There is nothing better than some Friday drama to close such a hectic week. A technical meeting on the #AI Act has broken down today, but make no mistake: the issue is deeply political and, if no solution is found soon, the whole law is at risk. A 🧵1/8 euractiv.com/section/artifi…

@PauseAI These arguments work just as well for non-AI-based attacks and defences. And yet financial institutions, companies, and governments are doing alright, even as the sophistication of attackers is ever increasing. Things will never be perfect, but there is no reason to believe AI… twitter.com/i/web/status/1…

New on arXiv!!! 🚨🚨🔥🔥🤯🤯 A Unified Framework to Enforce, Discover, and Promote Symmetry in Machine Learning by Sam Otto, with N. Zolman, N. Kutz, & myself A tour de force in machine learning & differential geometry arxiv.org/abs/2311.00212 (w/ beautiful drawings by Sam)

10^26 FLOPS ought to be enough for anyone

If AI can help the bad guys, it can also help the good guys protect against the bad guys. twitter.com/TacoCohen/stat…

In 1999 Apple's Power Mac G4 was considered a weapon by the US government. Why? Because it could perform over 1 gigaflop of calculations per second... Apple used this fact for its marketing.

@BlazejDolicki @MPontil @KatStankova It was a great event! Thanks for the invite 🙂

If AI actually gets good at finding security vulnerabilities, software will quickly become much more secure.

10^26 FLOPS ought to be enough for anyone

We are happy to announce that we are opening this year's ML in PL conference! We have four days ahead of us full of interesting presentations, inspiring panels and networking. See you on site!

Positional encoding is widely used in language but is suboptimal in 3D vision. We propose Geometric Transform Attention (GTA) that instead transforms Q, K, V into a shared coordinate system. GTA improves SotA Transformer methods on hard NVS benchmarks: arxiv.org/abs/2310.10375

We proudly present our 524 page book on equivariant convolutional networks. Coauthored by Patrick Forré, @erikverlinde and @wellingmax. maurice-weiler.gitlab.io/#cnn_book [1/N]