o3 is really special and everyone will need to update their intuition about what AI can/cannot do. while these are still early days, this system shows a genuine increase in intelligence, canaried by ARC-AGI semiprivate v1 scores: * GPT-2 (2019): 0% * GPT-3 (2020): 0% * GPT-4

Super excited to introduce QUICHE (Quantitative InterCellular Niche Enrichment) - a statistical method that can be used to discover local cellular niches differentially enriched in spatial regions, longitudinal samples, or clinical patient groups :) (1/10) biorxiv.org/content/10.110…

I attended last years version of the FOMO Winter School and I can't recommend it enough! When you're interested in Foundation Models (who's not?), this is it!

I’m super excited to share what I’ve been working on for the last (many) years: a spatial + genomic + transcriptomic characterization of how the breast cancer microenvironment evolves through immunotherapy! (1/x) biorxiv.org/content/10.110…

Making LLMs run efficiently can feel scary, but scaling isn’t magic, it’s math! We wanted to demystify the “systems view” of LLMs and wrote a little textbook called “How To Scale Your Model” which we’re releasing today. 1/n

Reinforcement Learning

Europe ♥️ Open Source.

o3 can't multiply beyond a few digits... But I think multiplication, addition, maze solving and easy-to-hard generalization is actually solvable on standard transformers... with recursive self-improvement. Below is the acc of a tiny model teaching itself how to add.

Tomorrow in our TUM AI - Lecture Series with none other than Robin Rombach (Robin Rombach), CEO Black Forest Labs. He'll talk about "𝐅𝐋𝐔𝐗: Flow Matching for Content Creation at Scale". Live stream: youtube.com/live/nrKKLJXBS… 6pm GMT+1 / 9am PST (Mon Feb 17rd)

Interactive segmentation with a nnU-Net? Better than all SAM-based interactive segmentation workflows? Whaaat!?

Non-invasive multiplexed microscopy imaging of living tissues at single cell resolution with ultrasound waves. This absolutely sounds like the next generation of tissue imaging and might have a huge impact on the field. Congrats David Maresca et al. for this achievement.

This is a big one! An active-learning workflow for cell phenotyping in multiplexed images which doesn't need any coding expertise. Download & install the stand-alone software and try it on your data.

Join the Spatial Biology Summit to get updated on recent trends in the field!

Thrilled to share that Nimbus, our method to robustly identify cell positivity in multiplexed imaging data, is out Nature Methods! If you've struggled with cell clustering/phenotyping in imaging data, please give it a try. nature.com/articles/s4159…

Nimbus predicts the likelihood of marker positivity from multiplexed image data in single cells. nature.com/articles/s4159…