finding
active
finding:direct-reduction-contributes-13-from-interventions-resolving-hallucinationsDirect reduction contributes ~13% from interventions resolving hallucinations
Decomposition component.
Source paper
extracted_from(2026) · Aaditya Vikram Prasad · Connor Watts · Jack Merullo · Dhruvil Gala +3
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Topline result.
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.
- The reduction in hallucination rate under inline interventions occurs due to a change in model beliefs in-context.hypothesis0.809Mechanism hypothesis for future work.
- Component of decomposition.
- 56% of test-time hallucinations are caught; of these 22% Fixed, 36% Retracted, 42% improperly handledfinding0.782Fig. 8 decomposition.
- Rules out overcautiousness.
- Implication of PRH: larger models should amplify bias less and hallucinate less if they better model reality
- Decomposition component.
- As models scale and converge toward an accurate model of reality, hallucinations should decrease with scalehypothesis0.755Implication of PRH for LLM hallucination
- Demonstrates practical utility of preventative steering in a realistic deployment scenario