claim
active
claim:character-training-is-more-robust-to-adversarial-prompting-than-activation-steering-on-averageCharacter training is more robust to adversarial prompting than activation steering on average
Robustness comparison claim; activation steering is brittle for QWEN 2.5 7B specifically
Source paper
extracted_from(2025) · Sharan Maiya · Henning Bartsch · Nathan Lambert · Evan Hubinger
Neighborhood — ranked by edge-count
Papers (1)
paper
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Character training is more robust to adversarial prompting than constraining system promptsclaim0.899Main robustness claim supported by classifier performance experiments
- Coherence advantage claim with mechanistic speculation about why steering leads to incoherence
- Realism advantage claim with safety implications; supported by anecdotal comparison in Appendix D
- Character training beats activation steering in coherence win rate 78.4% ± 5.2% on Llama 3.1 8Bfinding0.823Coherence comparison against steering baseline for Llama model
- Practical disadvantage of activation steering highlighted as a drawback
- Core claim distinguishing character training from role-playing, evidenced by robustness to break-character instructions
- Claim that models learn the spirit of the constitution, not just its letter, evidenced by suppression of opposing traits
- Character training changes the manner of responses rather than their informational contentclaim0.785Central claim demonstrated by Figure 1 where all responses constitute refusal but are conveyed differently