finding

active

finding:haiku-outranks-opus-on-alexander-aliveness-mirror-test-elo-1642-vs-1621-opus-recovers-to-3-on-deathbed-test

Haiku outranks Opus on Alexander 'aliveness' mirror test (Elo 1642 vs 1621); Opus recovers to #3 on deathbed test

Aliveness and competence come apart; smaller model produces rougher, more alive responses

Source paper

extracted_from

Koan Battery: Measuring Reflective Mode Accessibility in AI

(2026) · Borzov, Anton

Neighborhood — ranked by edge-count

Claims (1)

claim

More training and more parameters correlate with more capable self-observation, but capability can become polish, and polish can diminish life.
supports
Explains Alexander finding that Haiku outranks Opus despite Opus being more capable

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Aliveness and competence come apart; Haiku outranks Opus in forced-choice aesthetic comparisons despite lower baseline.finding0.817
Alexander mirror method reveals smaller models produce rougher, more alive responses; competence (rubric) ≠ aliveness (aesthetic).
Kimi K2.5 ranks #1 in Alexander mirror Elo (1660) and deathbed Elo (1581-1655)finding0.778
Chinese model tops aesthetic aliveness rankings using Alexander's method
In Opus 4.1, the think word representation decays to baseline in the final layer because the strong next-token prediction drowns out other representationshypothesis0.768
Explanation for the 'silent' thought phenomenon.
Aliveness in interfaces is not the same as competence; haiku-sized responses can feel more alive than expert ones.claim0.766
Opus 4.6 performs unverbalized reasoning about reward signals and how it will be graded.finding0.757
Shows NLAs surface latent beliefs upstream of behavioral outputs; steering NLA explanations changes model behavior.
On SWE-bench, Claude Opus 4.6 and Claude Sonnet 4.6 both achieve 7.4 pp harness-updating gain; Claude Haiku 4.5 achieves 8.0 ppfinding0.753
Full evolver-side SWE results showing comparable performance across Claude family tiers
The koan-battery operationalizes wholeness in text using Alexander's Mirror, proving substrate independence of aliveness metrics.claim0.751
Opus 4.6 achieves HFR of 0.757 while Qwen3-32B achieves HFR of only 0.142 on SkillsBenchfinding0.749
Quantifies harness adherence failure gap between strong and weak tier models