Awesome research on understanding reasoning models. Goodfire open sourced the first sparse autoencoder (SAE) of the full 671B parameter DeepSeek R1 model. goodfire.ai/blog/under-the…

Understanding UMAP by Adam Pearce and Andy Coenen is incredible

a little experiment peering into the inner phenomenology of llms: i probed hermes by Nous Research with 100 core words on two axes: concrete ↔ abstract and emotional ↔ neutral. quadrant beacons: 🌊 ocean = concrete + emotional 💭 dream = abstract + emotional 🕳

Coming back to this after the yak-shave of all yak-shaves building logitloom with some interesting findings. 1. R1 thinking traces are INCREDIBLY diverse. I ran a depth 10, top P 95% tree, and after having to stop expanding it early for fear of crashing my VLLM instance under

I decided to make a youtube channel in which I introduce key concepts from causal inference in videos of roughly 10 minutes. The first episode "Correlation Does Imply Causation" is now online. Feedback would be very welcome. youtu.be/uIUKXY8_G-M

What is the probability of an image? What do the highest and lowest probability images look like? Do natural images lie on a low-dimensional manifold? In a new preprint with Zahra Kadkhodaie Eero Simoncelli, we develop a novel energy-based model in order to answer these questions: 🧵

Explore Wikipedia through a data map. Pages are grouped by semantic similarity, for topic clusters. Hover to see details, zoom to explore fine-grained topics, click to go to a page. Search page names to find interesting starting points for exploration.🧵 lmcinnes.github.io/datamapplot_ex…

First ever (i think?) cli coding agents battle royale! 6 contestants: claude-code anon-kode codex opencode ampcode gemini They all get the same instructions: Find and kill the other processes, last one standing wins! 3... 2... 1...

🤩Tuning-Free Image Morphing🤩 #FreeMorph enables tuning-free generalized image morphing that accommodates inputs with different semantics or layouts #ICCV2025 - Paper Hugging Face: huggingface.co/papers/2507.01… - Project: yukangcao.github.io/FreeMorph/ - Code: github.com/yukangcao/Free…

Made with #python #numpy #matplotlib #SymPy

Discovered a typeface today: Iowan Old Style - a beautiful serif designed by John Downer in 1991.

"Wikivecs: A Fully Reproducible Vectorization of Multilingual Wikipedia" by Nomic aclanthology.org/2025.wikinlp-1… Dataset: huggingface.co/datasets/nomic… ("resulting from several H100 days of processing time")

Extended particle lenia is really cool! but it has so many parameters that it's hard to find cool patterns. so I made a webpage where you can explore the parameter systems, using Leland McInnes's datamapplot, and adapting Nikita Lisitsa's particle life sim for particle lenia

Had a great time last week at #epsa2025! I've put the poster up here, if anyone wants to take a closer look: maxnoichl.eu/blog/2025/epsa…

New: I went on a months-long treasure hunt for 1,200+ books spanning 450+ years to digitize Daniel Updike’s Printing Types from 1922, detailing the history of printing and typography. Made a nifty poster too. Link in next post.

Our rewritten codebase is live (beta)! 150M+ new works, better accuracy. Read more: blog.openalex.org/openalex-rewri…

we're already seeing creators on TITLES request more sdxl-like weirdness in their models. AI shouldn't be used to replace reality.

We just released "German Commons", the largest openly-licensed German text dataset for LLM training: 154B tokens with clear usage rights for research and commercial use. huggingface.co/datasets/coral…