Linear probe with a control task on a small transformer
Probe cached activations for a token-level property, build a matched control task, and report selectivity across layers and probe capacities.
Not attemptedSelf-directed project (control-task method)Access: free
The exercise happens on Self-directed project (control-task method). Canonical link: https://arxiv.org/abs/1909.03368. Link check: link responded (5 Oct 2026). Opened from this app 0 times; opening is not counted as an attempt.
Supports: Careful interpretability evidence for interpretability and research-ML roles.
Record an attempt
Attempt history
No attempts recorded yet.