David Bau: OV-transform summaries beat logit lens readouts in VLMs almost by accident

davidbau · x · 2026-09-30

Following up on the concept induction head research, davidbau highlights that students have been experimenting with "Jacobian lenses", and points to Sheridan Feucht's technique for summarizing attention-head-bundle OV transforms. Applied to VLMs, it produces readouts much better than the logit lens almost by accident.

Related event: Northeastern Researchers Unveil Dual-Route Model of Concept Induction Heads in LLMs(2 posts)→

Original post →

More from Research

Research channel →