Blog

Research notes, replications, and methodology lessons. Including the runs where we got it wrong the first time — those tend to be the most useful posts.

2026-04-2518 min readCaio Vicentino · OpenInterpretability
Entity-recognition features in Qwen3.6-27B — a replication, and a methodology lesson
AUROC 0.84 for the "I know this entity" feature on Qwen3.6-27B — vs Ferrando 2024's 0.73 on Gemma-2-2B-IT. Two-day arc with three controls: a tokenization confound that gave fake AUROC=1.0, single-feature steering that didn't move calibration, and multi-feature top-200 ablation that beats the random-K null at 4-8σ — but the LLM judge shows the "less hedging" is purely additional incorrect answers, not correct ones. We found a hallucination-induction mechanism, not a calibration knob.
SAEhallucinationQwen3.6-27BFerrando 2024methodologysteeringcircuitscontrols

More posts coming as research lands. RSS / mailing list when there are 5+.