Tweet by _chenglou

November 30, 2025

Explanations of Attention in ml are so complicated. I think of it as roughly a hash table lookup, except instead of using an exact key, you use a fuzzy one (query) to retrieve all related keys and average their values (A weighted average, where each weight is the query and that key’s affinity)

Author
_chenglou
Date
November 30, 2025