Jack Lindsey

Sources#

• Verbalizable Representations Form a Global Workspace in Language Models

Summary#

Entity. Researcher on Anthropic's interpretability team and corresponding author of Verbalizable Representations Form a Global Workspace in Language Models (Transformer Circuits, July 2026). With Wes Gurnee, conceived the Jacobian Lens (J-lens) and the conjecture linking verbalizable representations to conscious access.

Contributions#

Per the paper's author-contributions section:

• Conceived of the Jacobian Lens (J-lens) method and the verbalizability↔conscious-access connection (with Wes Gurnee)
• Ran the early experiments on the model's ability to directly modulate its own J-space (hold a concept in mind on instruction) and on the effects of post-training on lens readouts — the results that became The Assistant Persona in the Workspace
• With Nicholas Sofroniew, proposed the experiments connecting the J-space to global workspace theory — the move that turned an interpretability readout into a claim about the functional organization of model cognition

His prior transcoder/attribution work is also cited within the paper (the arithmetic features re-examined through the J-lens are from Lindsey et al.).

Connections#

• Jacobian Lens (J-lens) — co-originator
• The Global Workspace in Language Models (J-space) — corresponding author; proposed the workspace framing
• The Assistant Persona in the Workspace — ran the post-training diffing experiments
• Wes Gurnee — co-originator and co-first author
• Anthropic — interpretability team

Sources#

• Verbalizable Representations Form a Global Workspace in Language Models — Author Contributions; correspondence address