Artificial Intelligence · 25.08.2026, 12:02 UTC
What Does Activation Steering Control? Attribution Across Answer Encodings and Output-Sensitive Subspaces
| Schweregrad | info |
|---|---|
| Kategorie | Artificial Intelligence |
| Quelle | arXiv cs.CL ↗ |
| Veröffentlicht | 25.08.2026 UTC |
Sicherheitsmeldung mit Schweregrad noch nicht bewertet. Technische Details im Tab „Originaltext“; empfohlene Schritte in der Checkliste.
arXiv:2608.22985v1 Announce Type: new Abstract: Activation steering is often evaluated under the answer encoding used to construct the direction. A reported gain may reflect the intended judgment or compatibility with answer identifiers seen during construction. We introduce Cross-Encoding Steering Evaluation, which freezes an intervention while re-encoding answers to the same held-out items. On NormBank, after A/B/C identifiers are reassigned, contrastive activation addition (CAA) induces larger target-versus-source score changes for the extraction indices than for the semantic labels under the new mapping. We call this extraction-index following. Varying identifier vocabulary (A/B/C, X/Y/Z, or 1/2/3) and row order shows that the effect tracks extraction index rather than row position. After matching direction norms across layers, extraction-index following emerges mainly at later depths. A low-rank output-sensitive component containing 15.4% of the direction's squared norm retains 96.3% of this effect. An Inference-Time Intervention (ITI)-style method also favors extraction-index over semantic-label following on NormBank in three models. In aggregate, MNLI favors extraction-index following, whereas Social Chemistry 101 (SC101) favors semantic-label following. Multiple-choice and open-ended evaluations can yield different behavioral conclusions. Thus, a steering gain under one answer encoding does not by itself identify what the intervention controls.
Maßnahmen
⬇ Als MarkdownVerwandte Beiträge
- info Multi-Agent Orchestration with the Common-Sense Reasoning Capabilities of LLMs for Autonomous Driving
- info Decomposition Attacks Across Unlinkable Identities: Limits of Stateful Defenses for LLM Services
- info Towards Understanding On-Policy Distillation through the Lens of Test-Time Scaling
- info Do LLM Recommenders Know When They're Hallucinating? Auditing Confidence Calibration in Catalog Faithfulness