Hypothesis 2 (Persona Space): Persona vectors jointly compose a persona space; an instance's general dispositional profile is specified by a combination of activations along multiple persona vectors

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

• Hypothesis 3 (Persona Regions): There are stable regions (or basins of attraction) in persona space that correspond to coherent dispositional profileshypothesis0.872
  Third and most novel hypothesis; if confirmed, provides discrete individuation targets for both persona views
• There may exist a natural 'persona basis' characterizing the space of all model personas, with correlations between persona vectors predicting co-expression of traitshypothesis0.860
  Open question proposed by authors for future work on the dimensionality and structure of persona space
• Hypothesis 1 (Gateway Features): Persona vectors act as gateway features—single directions in activation space that shape LLM behavior across most, if not all, contextshypothesis0.839
  First of three hypotheses about persona implementation in LLMs, motivating the persona views
• The first principal component of persona space (the 'Assistant Axis') captures the assistant vs alternative persona distinction consistently across three models with role loading correlations exceeding 0.92finding0.823
  Finding establishing cross-model consistency of the assistant axis as the dominant structure in persona space
• A linear preference vector encoding how much a model likes a given task is persona-relative: it activates strongly for creative writing in the assistant but for phishing in the evil personafinding0.820
  Evidence that core representations like preferences are persona-relative, supporting claim that personas gate content of representations
• Persona vectors do not emerge as a single capability; each persona has its own onset and ceiling ordered by prevalence and behavioral simplicity in pretraining dataclaim0.805
  Interpretive finding against a unified emergence threshold for all personas
• Persona space is remarkably low-dimensional: just 4 components explain 70% of variance among 275 roles in Gemma 2 27Bclaim0.804
  Empirical support for Hypothesis 2, suggesting tractable structure in the space of LLM personas
• The directional transition pattern of persona vectors at the identified attention layer represents a unique footprint of persona representations, distinct from general hidden state dynamicsclaim0.803
  Supported by comparing persona vector transitions to hidden vector transitions from OpenAssistant data