Off-Topic Detector Latent Ablation

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Off-topic detector is a functional label based on selection methodology; these latents may serve broader coherence-monitoring roles beyond detecting off-topic contentclaim0.801
  Epistemic caution about over-interpreting the OTD label given the heterogeneity of identified latents
• Random Latent Ablation Controlmethod0.781
  Control experiment ablating random latents matched for activation frequency and magnitude to test OTD specificity
• Zero Ablationmethod0.760
  Intervention type that sets activations to zero, used for interpretability analysis
• 26 candidate off-topic detector latents identified in Llama-3.3-70B via contrastive searchfinding0.746
  Core mechanistic finding identifying specific SAE latents associated with ESR
• Feature ablation (zeroing feature activations)method0.722
  Clamping a feature's value to zero to measure its causal effect on model output.
• Boost Level Ablation Sweepmethod0.720
  Systematic sweep of 10 boost levels from threshold-3σ to threshold+3σ to characterize ESR vs. steering strength
• High-Low Frequency Detectorconcept0.720
  A less intuitive feature family detecting low-frequency patterns on one side of the receptive field and high-frequency on the other; used as example of non-obvious but understandable features
• Counterfactual Latent (CL) Lossframework0.716
  Auxiliary training objective from Grant (2025) that constrains intervened representations to remain near natural distribution