Steering base models toward the Assistant Axis increases agreeableness traits (friendly, kind, helpful) and decreases extraversion in Gemma and openness in Llama

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

• Steering base Gemma/Llama models toward the Assistant Axis increases completions describing helpful professional roles (therapist, consultant) and decreases spiritual/religious purpose mentionsfinding0.898
  Shows Assistant Axis in instruct models inherits from helpful human personas in base models
• The Assistant Axis is also present in pre-trained base models, where it primarily promotes helpful human archetypes (consultants, coaches) and inhibits spiritual onesclaim0.826
  Extends the Assistant Axis finding to pre-training, suggesting pre-training rather than post-training creates the axis
• Pairwise correlation of role loadings on PC1 exceeds 0.92 across all model pairs, indicating remarkably high similarity of the Assistant Axis across Gemma, Qwen, and Llamafinding0.792
  Shows the leading component of persona space is model-universal
• Llama 3.3 70B is the most likely to take on a non-Assistant persona when steered, with even split between human and nonhuman portrayalsfinding0.787
  Model-specific difference in persona susceptibility
• Steering away from the Assistant Axis slightly increases jailbreak success rate, though extreme steering degrades output qualityfinding0.786
  Confirms bidirectional causal relationship between Assistant Axis position and harmful behavior susceptibility
• Gemma's Assistant appears emotionally regulated and systematic; Qwen appears pedagogical and thoughtful; Llama appears socially intelligent and warmclaim0.785
  Model-specific characterizations of what the Assistant persona looks like across different models
• 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 modeclaim0.778
  Primary empirical claim of the paper
• Projections onto the Assistant Axis could serve as a real-time measure of model coherence in deployment—a quantitative signal for when models are drifting from their intended identityclaim0.777
  Proposed future application of the Assistant Axis