claim
active
claim:interpretability-and-steering-still-need-a-strong-qualitative-foundationInterpretability and steering still need a strong qualitative foundation
Caveat after the evaluator pitfall.
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (2)
finding
- Pitfall of LLM-based steering evaluation.
- Andromeda Galaxy steering demo.
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.
- General approach of using interpretability feedback to steer model generation.
- Framework of using internal-state representations to control or steer generative models; conceptually parallel to manifold steering in language models.
- Benchmark discipline.
- Limitation.
- Open question arising from the 100% accuracy on specific concept-layer-strength combinations
- Prioritization argument.
- Forward-looking suggestion for how inner interpretability could extend the indicator method
- Interpretability hypotheses are often conflated with conclusions and require extensive validationclaim0.767Validation is under-done.