new blog post "There Are No New Ideas In AI.... Only New Datasets" in which i summarize LLMs in exactly four breakthroughs and explain why it was really *data* all along that mattered... not algorithms

Small language models give me warm, fuzzy feelings! Google dropped this 270M model (Gemma), and it’s absolutely wild how coherent and fast it is on my phone. Locally installed, chatting without internet (PocketPal app).

doi.org/10.48550/arXiv…

Universal Learning of Nonlinear Dynamics One more cool paper on this. This one is mathematical but has a nice overview of existing methods for mere mortals. arxiv.org/abs/2508.11990

Please join us in congratulating HHMI Investigators Jennifer Doudna and David R. Liu (David R. Liu), recipients of national awards from the American Chemical Society (American Chemical Society) for 2026. 🔗cen.acs.org/people/awards/…

arxiv.org/html/2506.2173…

nature.com/articles/s4159…

MOSAIC adaptive optical lattice light sheet structured illumination microscopy mode 4D imaging and analysis of mitochondrial dynamics in a zebrafish embryo 14 hpf.

My lab has been pushing into explainable, robust, & theoretically-tractable AI models for science 💪 🚨At #ICCV2025 we introduce #DISTIL - led by amazing PhD student HOSSEIN - we propose a trigger-inversion method for DNNs that reconstructs malicious backdoor triggers 1/4

NVIDA chips are manufactured by TSMC, a Taiwanese company. They're created using EUV lithography machines manufactured by ASML, a Dutch company. These machines consist of >50% of German parts (by value), in particular ZEISS optics.

link.springer.com/article/10.100…

#神経科学トピックス 頭蓋骨透明化技術「SeeThrough（シースルー）」の開発 jnss.org/news-topics?id… doi.org/10.1038/s41467… Nature Communications 2025 Aug 26. SeeThrough: a rationally designed skull clearing technique for in vivo brain imaging

biorxiv.org/content/10.110… Jan Drugowitsch Interesting! Congrats

What if foundation models weren’t just for language & vision—but also for structured data? 📊🤖 Stable AI & Tsinghua University introduce LimiX, the first large structured-data model (LDM), designed to handle any tabular task through query-based conditional prediction. Key

Can AI help understand how the brain learns to see the world? Our latest study, led by Joséphine Raugel from FAIR at AI at Meta and École normale supérieure | PSL, is now out! 📄 arxiv.org/pdf/2508.18226 🧵 A thread:

Cooperative actions of interneuron families support the hippocampal spatial code science.org/doi/10.1126/sc…

Congratulations Chao Sun!!! 🎉

The most important AI paper of 2025 might have just dropped. NVIDIA lays out a framework for Small Language Model agents that could outcompete LLMs. Here’s the full breakdown (and why it matters):

So why use a massive LLM to do everything? Build modular agents: Use SLMs by default. Only call an LLM when truly necessary.

nature.com/articles/s4158…