Emergence of an alignment between LLMs' and the brain's computational dynamics, and key factors allowing it : scaling and context size.

We meta-learn a transformer based in-context learning fMRI visual cortex encoder which can adapt to new human subjects without any fine-tuning