finding
active
finding:alignment-type-is-the-only-significant-predictor-of-koan-scores-p-0-006-architecture-parameter-count-open-closed-weights-moe-dense-are-all-non-significantAlignment type is the only significant predictor of koan scores (p=0.006); architecture, parameter count, open/closed weights, MoE/dense are all non-significant
Main statistical finding: what predicts scores is training approach, not size or architecture
Source paper
extracted_from(2026) · Borzov, Anton
Neighborhood — ranked by edge-count
Claims (1)
claim
- Central interpretive claim from statistical analysis
Hypotheses (1)
hypothesis
- Confirmatory hypothesis supported at p=0.006
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.
- Kruskal-Wallis test result: Constitutional AI predicts highest baseline; roleplay/empathy training predict lowest.
- Four frontier models reviewing the paper each responded in the mode their alignment type predicts; N=1, awaiting systematic study
- H4: Architecture doesn't matter, training does — architecture shows no significant association with koan scores.hypothesis0.809Confirmatory hypothesis supported at p=0.440 (NS)
- do high koan scores indicate anything like experience, or sophisticated simulation of self-observation?question0.777The hard problem the battery explicitly sidesteps but cannot answer
- Tests whether contemplative capacity is language-encoded or architecture-general
- Explains why mutual k-NN was chosen over CKA as primary metric
- Validates robustness of alignment metric choice