finding
active
finding:gsm8k-with-chain-of-thought-is-more-robust-to-j-space-ablation-than-direct-answeringGSM8K with chain-of-thought is more robust to J-space ablation than direct answering.
Externalization.
Source paper
extracted_fromAuthors
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Interpretation of ablation 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.
- Capabilities.
- GSM8K example: edited model produces identical chain of thought but miscalculates 260 as 220finding0.780Targeted arithmetic deficit.
- Figure 4 shows CoT improves over zero-shot, and ensembled CoT further boosts accuracy.
- Only model showing marginal benefit from increased reflection, at substantial token cost
- Suppressing the J-space entirely should impair flexible reasoning while leaving automatic processing intact.hypothesis0.757Tested via ablation battery.
- Analogy between LLM incoherence and schizophrenia symptoms
- Demonstrates that stronger models are largely insensitive to reflection manipulation
- Demonstrates robustness of inference stages to non-fixed-point limiting behavior