Comparison of Inoculation Prompting (IP) and Counterfactual Reflection Training (CRT) for sycophancy in Qwen3-8B, with behavioral and mechanistic (steering, patching) analysis. IP behaved like a gate (easily re-elicited); CRT behaved like a broader rewrite (harder to restore but overcorrected).
Companion codeCase study on why AI coding agents need eval-first engineering and how coding agents should code agents.
Public writeup on wording confounds in contrastive activation extraction across 6 traits and 2 open-weight models. Companion to the Deconfounding Contrastive Interpretability project.
Companion codePublic writeup on mechanism-specific limits of introspective tamper detection in Qwen3-8B. Companion to the Neuron Suppression Awareness project.
Companion code"Make it work. Make it right. Make it fast. In that order."