OTIP-Lite: A Minimal Protocol for Detecting Stale Continuation, Sham Overreaction, and Purpose Leakage Under Intervention
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Outcome-level success can overstate capability when a model produces acceptable answers through smooth continuation, cosmetic compliance, or direction-insensitive response behavior rather than through a genuine update in governing task orientation. OTIP-Lite: A Minimal Protocol for Detecting Stale Continuation, Sham Overreaction, and Purpose Leakage Under Intervention introduces a low-resource admissibility protocol for auditing whether a model’s output is actually governed by a valid intervention. OTIP-Lite is not presented as a general benchmark, a theory of consciousness, or a full alignment framework. Instead, it is a compact audit layer organized around three gates: Sham Separation, Directionality/Sign-Flip, and Persistence After Valid Update. The note also proposes an additional refinement, purpose leakage, to capture cases in which a model mirrors the requested update at the content level while preserving the prior communicative purpose beneath the new surface. The manuscript is positioned as a methodological extension in continuity with the broader DDG-OTIP line. Its present contribution is not benchmark-scale empirical validation, but a bounded protocol for distinguishing apparent success from maintained orientation under intervention in low-resource settings. A worked dry-run is included only as a non-admissible illustration of protocol logic, not as benchmark evidence.
Fields of Study & Subjects
Ethics and Social Impacts of AI
Explainable Artificial Intelligence (XAI)
Topic Modeling
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 30 March 2026
- Language
- EN