concept
active
concept:misfire-of-known-answer-circuitMisfire of known-answer circuit
Hallucinations from known-entity features firing without real knowledge.
Neighborhood — ranked by edge-count
Papers (1)
paper
Thinkers (1)
thinker
- Ferrando et al.studiesFound known/unknown entity features with SAEs.
Claims (1)
claim
- Karpathy case.
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.
- Lists issues: concept-based tasks, flawed decomposition, low faithfulness.
- Interpretability technique for identifying functional sub-circuits in neural networks, supported by pyvene
- Circuits active by default, e.g. can't-answer, suppressed by contrary evidence.
- Operational misfits in concepts reveal design flaws without requiring implementation or testinghypothesis0.684Jackson's hypothesis that negative scenarios can identify conceptual design problems early
- Removing duplicate persistent features to shrink circuit depth.
- Metacognition caveat.
- Anthropic technique revealing computational pathways in Claude.
- Prior mechanistic interpretability work reverse-engineering vision models (InceptionV1); the direct predecessor this paper extends to language models