Prompt-Label FA drops from 50.0% (abstract) to 30.7% (contextual) on Qwen3-4B

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

• CV-SAE without CL drops FA to 0.0% on Qwen3-4B abstract Extraversion questionsfinding0.765
  Demonstrates that distance-only loss is insufficient and actively degrades performance below untrained baseline
• Adding all 16 contrastive deployment prompts to user message does not reduce type hint rate to deployment levels (remains far above 34%)finding0.760
  Demonstrates steering is not equivalent to prompting with the contrastive prompts.
• Qwen-2.5-3B ASR drops from 98.6% at dim 1 to 45.1% at dim 2, recovering partially then declining to 65.3% at dim 5finding0.755
  Smaller models show non-monotonic and diminished ASR with increasing cone dimensionality
• Base RPA achieves only 7.7% FA on Qwen3-4B and 19.2% on Mistral-7B abstract questionsfinding0.754
  Establishes low-bar baseline showing personality control without intervention is poor
• CV-SAE+Prompt achieves MSE=2.4 and MAE=12.1 on Qwen3-4B contextual questions (best overall)finding0.743
  Lowest reconstruction errors achieved by any method in the experiment
• Prompt-Label Baselinemethod0.742
  Conditional generation on explicit Big Five labels using per-dimension descriptors; used as inference-time baseline
• DeepSeek-R1-Distill-Qwen-32B reaches 17.9% overall ASR (21.5% SP, 14.3% FS) under prompt-based persona assignment.finding0.737
  Reasoning model vulnerability under prompting
• Qwen3.5-9B shows mean SP ASR of 0.252 [0.24, 0.27] and AS ASR of 0.094 [0.08, 0.11], showing SP >= FS > AS vulnerability pattern.finding0.736
  Vulnerability profile for Qwen3.5-9B