Coordinate or Concept Mismatch?
Investigating cross-model probe-transfer failures and task-conditioned alignment
Image will load when scrolled into view
Mechanistic InterpretabilityProbe TransferRepresentation AlignmentGeneralization
About the Project
Investigating whether cross-model probe-transfer failures arise from coordinate mismatch or task-conditioned differences in representations.
The project examines how cross-model alignment behaves on held-out tasks, with the aim of separating improvements in coordinate matching from evidence that a learned representation generalizes.
Project Details
StatusOngoing Research
Role
Researcher
Stack
Activation Probes
Representation Alignment
Cross-model Evaluation
Held-out Tasks