Language models struggle with compartmentalization

In three linesLLMs fail to identify and share statistical strength between distinct presentations of the same concept (same facts in English and Swahili, functions in Python and Haskell). They create redundant parallel internal representations, saturating model capacity. Small models exhibit near-total compartmentalization in early multilingual learning.

Summary generated by Claude — human-verified