← HR & people

Causal Tracing

Locates where an AI model stores a fact by corrupting and restoring its internals.

/causal-tracingndif-team/skills

More in HR & people

  • AblationTests what a model component contributes by removing it.
  • Attention AnalysisAnalyzes attention patterns and per-head behavior in transformer models.
  • Carta Compensation ScorecardCompares employee compensation to market.
  • Carta Fund Admin RequestsSends tasks to your Carta fund admin team and tracks their replies.
  • ProbingTrains probes on model activations to test what information a model encodes.
  • Plan ArchiveArchive a completed or abandoned plan for future reference.
  • Replication DesignerDesigns a replication of a published study.
  • TasksCreates, inspects, and comments on Sokosumi coworker tasks through the headless CLI.