LLMs excel at fitting finetuning data, but are they learning to reason or just parroting🦜? We found a way to probe a model's learning process to reveal *how* each example is learned. This lets us predict model generalization using only training data, amongst other insights: 🧵

🚨Announcing our #ICLR2025 Oral! 🔥Diffusion LMs are on the rise for parallel text generation! But unlike autoregressive LMs, they struggle with quality, fixed-length constraints & lack of KV caching. 🚀Introducing Block Diffusion—combining autoregressive and diffusion models

Excited to introduce funROI: A Python package for functional ROI analyses of fMRI data! funroi.readthedocs.io/en/latest/ #fMRI #Neuroimaging #Python #OpenScience Work w Anna Ivanova 🧵👇

Today, we're releasing a new paper – One-Minute Video Generation with Test-Time Training. We add TTT layers to a pre-trained Transformer and fine-tune it to generate one-minute Tom and Jerry cartoons with strong temporal consistency. Every video below is produced directly by

For those attending NAACL, today I'll be presenting recent work on how we can use language encoding models to identify functional specialization throughout cortex. Stop by my talk at 10:30 at the CMCL workshop!

🤯Your LLM just threw away 99.9 % of what it knows. Standard decoding samples one token at a time and discards the rest of the probability mass. Mixture of Inputs (MoI) rescues that lost information, feeding it back for more nuanced expressions. It is a brand new

We'll be presenting this at #ACL2025 ! Come find me and Thomas Jiralerspong in Vienna :)

Can we precisely and noninvasively modulate deep brain activity just by riding the natural visual feed? 👁️🧠 In our new preprint, we use brain models to craft subtle image changes that steer deep neural populations in primate IT cortex. Just pixels. 📝arxiv.org/abs/2506.05633

New paper: Ask 35 simple questions about sentences in a story and use the answers to predict brain responses. Interpretable. Compact. Surprisingly high performance in both fMRI and ECoG. biorxiv.org/content/10.110…

Richard Antonello Very cool video. I created an infographic to try to visualize the full study in more depth studyvisuals.com/artificial-int…

*Harnessing the Universal Geometry of Embeddings* by Rishi Jha Jack Morris Vitaly Shmatikov With the proper set of losses, text embeddings from different models can be aligned with no paired data (what they call the "strong" Platonic hypothesis). arxiv.org/abs/2505.12540

If you're at #SNL2025 and curious about speech and music perception or its representation in developing brains, stop by my poster in Session C – #77! :)

Human auditory cortex integrates information in speech across absolute time (e.g., 200 ms), not phonemes, syllables, words, or any other time-varying speech structure: nature.com/articles/s4159…

As our lab started to build encoding 🧠 models, we were trying to figure out best practices in the field. So Taha Binhuraib 🦉 built a library to easily compare design choices & model features across datasets! We hope it will be useful to the community & plan to keep expanding it! 1/