Simon Maris

📣 TRANSFORM 2023 📣 Conference on AI, Sustainability, Art and Design. 25 – 28 October 2023 Campus of Art and Design University of Applied Sciences Trier, Germany Program and registration at: kind-lab.de/transform2023 #ai #sustainability #art #design #transform23 #kindlab #trier

DeepSeek R1 latent visualization: AME(R1)CA (AMERICAN_FEATURE)

Die Realität hat das #ZDf als auch @DasErste eingeholt: Die Zahlen sagen klar, dass es ein #TriellMitHabeck werden muss. Von meiner Seite noch: Es muss auch direkt im Fernsehen #barrierefrei mit #Gebärdensprache und #Untertitel sein. #Inklusion #eineechteAuswahl

Modern presidential elections are marked by striking electoral stability.

Today the symposium continues with @tobias_revell, who's blog and work I have been following for years! Great talk, thank you, Tobias! Looking forward to another intense day and visits around DDW in the evening and tomorrow morning, let me know if you are in Eindhoven as well!

Yesterday I gave a short talk about our research project @gestaltungai on the integration of AI in design education at the Design and AI Symposium at TU/e, wonderful session hosted by Stephan Wensveen with great input by Janet Huang and @davemurrayrust and a dedicated audience!

@lkwq007 @adhik_Joshi OK thanks for the reply and don’t work too hard! Looking forward to it and observing GitHub for now.. sounds promising

@simonmaris @adhik_Joshi I’ve been a bit busy lately, so I’m unsure about the ETA. The features implemented but not yet released include better TPU support (~9.9s/sample at 1024x1024 for 50-step FLUX-dev on a TPUv4-8, but FLOPS utilization is only ~30%), sharding support, and some Pallas kernels for FLUX

#FLUX JAX-based 1-step FLUX.1-schnell generates a 1024x1024 image in 0.4s; 2-step takes ~0.6s, and 4-step takes ~1.3s.

@lkwq007 @adhik_Joshi would love to see your more optimized version, the current code runs great for me. was thinking about adding a few things but it seems you are ahead? care to give an ETA? :)

@adhik_Joshi No model-specific optimizations have been implemented yet. The demo is primarily based on my JAX port of Flux: github.com/lkwq007/flux-f… . A more optimized version (used by the demo) will be released soon.

The conference on Artificial Intelligence & Fundamental Rights kicked off yesterday with a focus on the impact of the European #AIAct. This was the refreshing first session with @j2bryson, looking forward to today's programme. Great to have this in Trier, thank you @IRDTrier!

Protein designers should be designing, not pipetting. AI models like AlphaFold, RFdiffusion & co make it easy to design new proteins. But a protein on the computer isn’t worth much by itself. At @adaptyvbio we’re building a platform that allows you to actually test your AI-designed proteins in the real-world! Upload your sequences and let our fully automated lab run the experiments. Then, download your results and feed them back into the AI models to design better proteins. Get started now: adaptyvbio.com/blog/beta

Excited to release a research preview of Feature Lab, a playground where you can sculpt images by editing their features. It uses a SAE to extract the sparse features from an image embeddings model, and allows us to finely sculpt and tweak images.

Using diffusion models to generate spectrograms that look like images and can also be played as sound. Reminds me to my diploma project on steroids. ificl.github.io/images-that-so…

So here's a story of, by far, the weirdest bug I've encountered in my CS career. Along with @maciejwolczyk we've been training a neural network that learns how to play NetHack, an old roguelike game, that looks like in the screenshot. Recenlty, something unexpected happened.

Meanwhile, over in Google Search. Andrew Johnson has been killin it, I never knew.

🌍 The workshop "AI Ethics from the Majority World: Reconstructing the Global Debate through Decolonial Lenses" has just started at @iwe_bonn @UniBonn! Prof Dr van Wynsberghe @aimeevanrobot opened the event with an inspiring speech, setting the stage for discussions…1/2

Feeling the AGI stabilityagi.com

swollen LiPo on a remote island. the ferry comes once a year and has no space for e-waste. 32C/90F everyday, soon it may explode—every building is a flammable thatched hut. “what if we bury it”, then the lithium goes into the ground water? welcome to the end of the supply chain.

@SashaMTL i was considering doing (a part) of our recent conference on ai, sustainability and design in a deconsecrated church, in the end it didn't work out, maybe 2025?

God is dead, long live ChatGPT 💀

We are thrilled to present a new #enzyme to the world, the first shown to break the man-made carbon-silicon bonds in #siloxanes, which are persistent in the environment. caltech.edu/about/news/tea…

Fine-grained control/editing in text-to-music diffusion models w/NO TRAINING? Presenting DITTO: Diffusion Inference-Time T-Optimization for Music Generation 📖:arxiv.org/abs/2401.12179 🎹:ditto-music.github.io/web/ w/@McAuleyLabUCSD @BergKirkpatrick @NicholasJBryan🧵