the.ai

Interpretability / Methods

verified

Probing Classifier

Train a small classifier on a model's internal activations to predict some property — part of speech, sentiment, whether a position is winning. If it succeeds, the property is available in there. The catch, which took the field a while to take seriously, is that a powerful enough probe can learn the property itself rather than find it.

Viz primitive · loss-curvesteps = 1000 · lr = 0.002 · batch = 64 · params = 1
loss
step 0dashed = held-out1000

Loss over 1000 training steps, starting near 7.2. It falls to about 1.95, with 93% of the total improvement arriving in the first half. A second line shows held-out, ending higher at about 2.35.

A probe's loss against the same probe measured on data it did not fit. Drag the probe capacity up to watch it drive the fitted curve down while the held-out line trails it — that gap is what separates a probe that read the representation from one that learned the task.

1

Reviewed by opendroid · 2026-08-18