finding
active
finding:nla-exposes-memorized-association-for-a5-case-moving-probability-0-25-to-0-72NLA exposes memorized association for *&a5 case, moving probability 0.25 to 0.72
Qualitative win.
Source paper
extracted_from(2026) · Adam Karvonen · Euan Ong · Subhash Kantamneni · Samuel Marks
Neighborhood — ranked by edge-count
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.
- Soft clues redundant with transcript.
- Tool-discovered hypotheses are easily guessed.
- Key limitation identified: NLAs hallucinate specific details while preserving thematic accuracy; informs practical usage.
- Initial evidence that alignment faking persona is more sensitive to exploiting training signals
- Appendix B.
- Tool-induced answer flips mostly toward 'no effect' (NLA 82% positive, 84% negative), cancelling in accuracyfinding0.736Direction of tool-driven prediction changes.
- Core limitation and usage heuristic: read NLAs for themes rather than individual factual claims; cross-check with original context.
- Quantitative evidence that NLA training produces increasingly informative explanations despite optimizing only for reconstruction.