finding
active
finding:56-of-test-time-hallucinations-are-caught-of-these-22-fixed-36-retracted-42-improperly-handled56% of test-time hallucinations are caught; of these 22% Fixed, 36% Retracted, 42% improperly handled
Fig. 8 decomposition.
Source paper
extracted_from(2026) · Aaditya Vikram Prasad · Connor Watts · Jack Merullo · Dhruvil Gala +3
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Model features can serve as inexpensive, dense, scalable supervision for open-ended tasks.associated_withCentral thesis.
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.
- 60.7% of hallucination mistakes corrected by adding vision features in two-stage framework on ScienceQAfinding0.815Quantitative evidence that vision information mitigates hallucinated rationales; 56% of error cases contained hallucinations, 60.7% of which were resolved with vision features.
- Decomposition component.
- Karpathy case.
- Primary negative result of the study: temporal permutation analysis finds no statistically significant indicators of consciousness in LLM representations.
- Rules out overcautiousness.
- The reduction in hallucination rate under inline interventions occurs due to a change in model beliefs in-context.hypothesis0.755Mechanism hypothesis for future work.
- Predictive hypothesis driving the investigation in Section 3.3; supported by experimental evidence.
- Demonstration of failure mode of abductive model reduction