finding

active

finding:trait-space-requires-4-dimensions-gemma-qwen-and-7-dimensions-llama-to-explain-70-of-variance-with-distinctive-pc1-spanning-conscientious-to-impulsive-traits

Trait space requires 4 dimensions (Gemma, Qwen) and 7 dimensions (Llama) to explain 70% of variance, with distinctive PC1 spanning conscientious to impulsive traits

Corroborates role space findings using traits; shows PC1 also captures Assistant-ness in trait space

Source paper

extracted_from

The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models

(2026) · Christina Lu · Jack Gallagher · Jonathan Michala · Kyle Fish +1

Neighborhood — ranked by edge-count

Claims (1)

claim

The leading component of the persona space of instruct LLMs is an 'Assistant Axis' that captures the extent to which a model is operating in its default Assistant mode
supports
Primary empirical claim of the paper

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.

The difficulty boundary for truth directions replicates across all four tested models (Llama-3.2-3B, Llama-3.1-8B, Gemma-2-2b, Gemma-2-9b); generalization to F3-F5 remains consistently low regardless of model size or family.finding0.781
Establishes generalizability of the core difficulty-boundary finding across model families.
4-19 principal components explain 70% of variance in role persona space across the three models (Gemma 4, Qwen 8, Llama 19)finding0.772
Demonstrates that persona space is low-dimensional
Introspective capacity scales with model size for some concepts, approaching near-perfect coupling in LLaMA-3.1-8Bclaim0.771
Validated for wellbeing and interest; focus and impulsivity do not show consistent scaling
LLaMA-3.2-1B impulsivity introspection: ρ=0.21, p<10⁻⁴ (significant but weaker than 3B ρ=0.52)finding0.761
Impulsivity shows significant introspection in 1B but declines in 8B; non-monotonic scaling
Most independent dimension pair is aesthetic_response and boundary_awareness (rho=0.553); most correlated is prediction_error and conceptual_crystallization (rho=0.886)finding0.760
Characterizes internal structure of the six scoring dimensions
In LLaMA-2-7B, PCA of larger_than+smaller_than shows statements clustering by surface-level characteristics (e.g., presence of token 'eighty') rather than truth valuefinding0.752
Shows absence of abstract truth representations in smallest model, supporting scale-dependent emergence claim
Persona space components explain 19.4%-33.6% of overall activation variance on LMSYS-CHAT-1M across the three modelsfinding0.752
Shows persona space captures a substantial portion of real conversational activation variance
Pairwise similarity of trait PC1 across all three models is >0.81; no pairwise correlation in top 3 trait PCs is below 0.70finding0.749
Shows trait space has more cross-model consistency than role space beyond PC1