CV-SAE without CL drops FA to 0.0% on Qwen3-4B abstract Extraversion questions

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

• CV-SAE with CL achieves 76.9% FA on Qwen3-4B abstract Extraversion questions (up from 11.5% before training)finding0.899
  Demonstrates the critical contribution of contrastive learning to control vector alignment
• CV methods were resilient and often improved under contextualized questions; CV-SAE+Prompt gained +7.5pp FA on Qwen3-4Bfinding0.797
  Demonstrates that latent steering generalizes better to situational cues than prompt-only methods
• CV-SAE achieves 7.0% MTR on Mistral-7B abstract questions vs 29.1% for CV-CAAfinding0.779
  Demonstrates that SAE-based injection is substantially more stable than CAA on Mistral-7B
• CV-SAE+Prompt achieves MSE=2.4 and MAE=12.1 on Qwen3-4B contextual questions (best overall)finding0.778
  Lowest reconstruction errors achieved by any method in the experiment
• Both CV-SAE and CV-CAA substantially outperform Prompt-Label baseline across FA, MSE, and MAE on both backbonesfinding0.776
  Confirms effectiveness of direct vector modulation over prompt-only conditioning
• Mid-range α values with CV-SAE deliver high FA and low errors, while CV-CAA peaks at smaller α and shows faster MTR growthfinding0.773
  Characterizes the differential sensitivity to injection strength between SAE and CAA methods
• CV-SAE provides a more concentrated 'purer' steering direction with less extraneous content than CV-CAA, enabling higher FA without rising MTRclaim0.769
  Author's interpretation of why SAE outperforms CAA in stability at higher injection strengths
• Prompt-Label FA drops from 50.0% (abstract) to 30.7% (contextual) on Qwen3-4Bfinding0.765
  Shows that explicit labels without latent steering fail to generalize to situational cues