From attention to causal evidence

Year 12 ยท Student investigation

Can we recover an answer by restoring one internal state?

Keep an individual record even when you investigate in pairs. Record the exact settings, source and data split.

Predict before seeing the answer

Will patching the final token after block 1 fully recover the clean logit difference? Why might an earlier patch behave differently?

Two comparisons

What did you change? What stayed the same? What did you notice?

Challenge your first explanation

Compare final-token patching after the final block with a patch at an earlier token. Then repeat on a different card.

Make and explain

A matched intervention report with clean, corrupted and intervened scores, recovery and scope.

Limit the claim

This activity can show _ . It cannot tell us _ . A new case I would test is ____ .

Use fictional data. Download or print your work before leaving.