finding

active

finding:only-46-15-of-cases-show-covariance-patterns-consistent-with-the-big-two-model-no-llm-satisfies-all-big-two-correlations

Only 46.15% of cases show covariance patterns consistent with the Big Two model; no LLM satisfies all Big Two correlations

Suggests a gap between LLM learned representations and human personality structure as described by Big Two

Source paper

extracted_from

Psychological Steering of Large Language Models

(2026) · Leonardo Blas · Robin Jia · Emilio Ferrara

Neighborhood — ranked by edge-count

Claims (1)

claim

OCEAN MDS injection covariance patterns departing from the Big Two model suggest a gap between learned LLM representations and human psychology
supports
Interpretive conclusion from Big Two mismatch finding; tentative due to only 46.15% match rate

Frameworks (1)

framework

Big Two Model
contradicts
Meta-trait model grouping OCEAN traits into stability (C, A, reversed N) and plasticity (E, O); used to evaluate covariance patterns from injections

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.

Agreeableness-Conscientiousness Big Two correlation is observed in 10 of 13 LLMs; N-C correlation is rarest (1 LLM)finding0.798
Most and least common Big Two covariance pattern in LLM OCEAN MDS injections
Models with as few as 14B parameters can exhibit 40% misalignment while maintaining 99% coherence (Turner et al. 2025 result)finding0.776
Concurrent work result showing emergent misalignment occurs in small models
Li et al. 2024: larger LLMs outperform smaller ones at distinguishing self-related from non-self-related properties on self-awareness benchmarksfinding0.769
Prior finding showing scale-dependent self-awareness, consistent with the scale effect observed in the paper's Experiment 1
The consistency with which LMs exhibit character traits varies with model size, fine-tuning, and prompting.claim0.764
Finding replicated across multiple experiments.
Models fine-tuned on human data show strong positive correlation between misalignment and incoherence; misalignment scores fall below 1% when incoherence threshold is appliedfinding0.763
Human data fine-tuning effect is distinct from synthetic emergent misalignment and likely caused by off-policy training
Would human raters and cross-judge replication corroborate the LLM-as-judge assessments of coherence and trait expression?question0.760
Methodological concern raised about potential bias and circularity of model-based classifiers
Overall human-LLM judge agreement rate for coherency is 91.7% across 120 pairwise judgmentsfinding0.755
Validates the LLM-as-a-Judge evaluation protocol for coherency scoring
All cases satisfying Criteria 1 and 2 (two out of three) originate from deeper transformer layers and/or the 2/3 layer of LLMs.finding0.753
Consistent with literature that deeper layers encode semantic information and align with human brain activity.