Pinned
we (acsresearch.org) expanded this into a larger paper! (my first.) we added some new experiments and found an interesting correlation - prompts that encourage the model to say there is an injection, even when there isn't one, correlate with better concept identification! arxiv.org/abs/2602.20031
new blog post! can small, open-source models also introspect, detecting when foreign concepts have been injected into their activations? yes! (thread, or full post here: vgel.me/posts/qwen-i...)