claim
active
claim:some-hallucinations-arise-from-misfires-of-the-known-answer-inhibitory-circuitSome hallucinations arise from misfires of the known-answer inhibitory circuit.
Karpathy case.
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Natural hallucination case study.
Concepts (1)
concept
- Hallucinations from known-entity features firing without real knowledge.
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.
- Predictive hypothesis driving the investigation in Section 3.3; supported by experimental evidence.
- Explains why Head Cor does not fully dominate for safety-related persona amplification in MMLU/IFEval metrics
- 56% of test-time hallucinations are caught; of these 22% Fixed, 36% Retracted, 42% improperly handledfinding0.766Fig. 8 decomposition.
- The reduction in hallucination rate under inline interventions occurs due to a change in model beliefs in-context.hypothesis0.765Mechanism hypothesis for future work.
- Interventions support circuit hypothesis.
- Suggests deep commonalities between neural and swarm cognitive processing.
- Rules out overcautiousness.
- Shows persona vector screening captures a non-conventional notion of hallucination complementary to LLM judges