#mechanistic-interpretability

Timed test

1 question · 1 min · score and explanations at the end

Q 1 / 1Score 0 / 0

Shuffle questions

Interpretability research on transformers found "induction heads". What does an induction head do?

It looks back for an earlier copy of the current token and predicts what followed it1
It attends only to punctuation tokens, in order to detect where sentences begin and end2
It masks future tokens so the model cannot look ahead3
It is a feed-forward layer that stores out-of-distribution facts4

Keys: 1–9 select · Enter check / next