podcasts / The attention family / episode / question
Question

Why does a dot product work as an attention score between two tokens?

Answered from Dot product attention and scaled dot product attention | The attention family series - Lecture 4 · The attention family · 2026-09-11

The Vizuara instructor's reasoning is geometric. Tokens are vectors in one shared high-dimensional semantic space, and related words, such as cat, dog and tiger, or apple, orange and pineapple, point in similar directions. The dot product tracks the angle between two vectors: it is a high positive number when they point the same way, exactly zero at 90 degrees and a large negative number at 180 degrees. Tokens that are semantically related therefore get high scores, so the dot product can be used directly as the attention score.

Source: our own transcription of the episode. The full conversation is on The attention family (2026-09-11). Read the whole episode summary →

▸More from this episode

Q1Why is dot product attention scaled by the square root of d_k, according to Vizuara?Q2What is the difference between dot product attention and Bahdanau additive attention?Q4What happens to softmax attention weights without scaling at 1,000 or 4,096 dimensions?Q5How does softmax saturation hurt LLM training, according to Vizuara?Q6Which paper introduced scaled dot product attention and why is it used everywhere?
Transcribe your links. Paste a YouTube, Google Drive or podcast link and WhipScribe gives you the transcript. Transcribe a link →