🤖📈 How do LLMs use their depth? Akshat Gupta led a fun project to find out! We leverage TunedLens (~linear decoding of tokens) to explore how LLMs' internal representations change from layer to layer. Preprint: arxiv.org/abs/2510.18871 1/

The statistical frameworks we introduced- DIP and PIP- go beyond audio&video, they’re general tools for quantifying how two feature-based models differ or converge in predicting a common target over time, revealing patterns of dominance and joint encoding across brains🧠, models🤖, and modalities🎬🎧.

We uncover two complementary cortical mechanisms underlying how the brain processes naturalistic audiovisual information - one system flexibly switches what sensory modality it encodes; another one jointly encodes auditory + visual inputs. ⚡ Quick thread with the main findings 👇

Excited to share our work on mechanisms of naturalistic audiovisual processing in the human brain 🧠🎬!! www.biorxiv.org/content/10.1...