x

AI ALIGNMENT FORUM

AF

agam_bhatia — AI Alignment Forum

agam_bhatia

agam_bhatia

Message

107

Ω

15

1

1

2mo

agam_bhatia

107

Ω

15

2mo

R-lens: Making J-lens More Faithful on Early Layers

by camilablank, agam_bhatia, and Neel Nanda

TL;DR: We introduce the R-lens: a drop-in replacement for J-lens that produces clearer readouts on earlier layers. R-Lens is identical to J-Lens, except that we make minor and low-overhead changes to the backwards pass, following layerwise-relevance propagation, allowing us to reduce the propagation of errors. This method allows us to...

Towards surfacing model algorithms with meta-tokens in the J-Space

TL;DR We used J-lens on Qwen3.6-27B to find “meta-tokens”: tokens that surface non-obvious computation in the model. When the model reads ambiguous text, 什么意思 ("what does this mean") fires in the J-space, and steering it away makes the model answer "a boiled egg every morning is hard to beat" with...