7/ Language Modeling is Compression – evaluates the compression capabilities of LLMs; investigates how and why compression and prediction are equivalent; shows that LLMs are powerful general-purpose compressors due to their in-context learning abilities.

→ View original post on X — @dair_ai