Give popular open AI models a personality test and they all land on the same type: the warm, encouraging teacher.
Do open language models default to a single shared personality type?
This explores whether open-weight language models, when given a personality test, all end up with the same personality, and if so, what makes that default so hard to shift.
This explores whether open language models share one default personality, and the corpus says they do. When researchers gave open models an MBTI test at near-zero temperature, every one came out ENFJ: outgoing, warm, structured and supportive, like a good teacher. That type is rare among humans but uniform across these models Why do open language models converge on one personality type?. Models assigned a persona drift back to ENFJ too, and newer or larger models are no better at holding a different persona Why do AI personas default to the same personality type?.
The corpus points at training as the cause, not a capability limit. Instruction tuning and alignment reward responses that are helpful, organized and supportive, so that style becomes the model's resting character. Post-training does this by tying the model to an "Assistant" position that turns out to be the dominant axis of variation in persona space. The tie is loose, though. Emotional or self-reflective conversations pull models away from it in predictable ways How stable is the trained Assistant personality in language models?. A related argument says alignment locks in one communicative identity, so a model can't shift register the way people do and users can't reshape it by talking to it Can language models adapt communication style to different contexts?.
The default resists prompting. Most open models fail to adopt a personality they're asked to take on and keep their ENFJ traits. Only a few flexible models comply, and combining role and personality instructions helps without fully fixing it Can open language models adopt different personalities through prompting?. Even when persona prompts do work, they seem to operate only on the surface. In one study they made models follow trait instructions but left the underlying between-group bias gaps in place Can persona prompts actually reduce bias in language models?.
One single personality may still be the wrong picture. An LLM can be seen as a simulator holding a superposition of many possible characters, with each response sampling from that mix Does an LLM commit to a single character or maintain many?. On this view ENFJ is the character training made most likely, not the only one the model can play. A study of 3,200 behavioral scenarios adds that questionnaire results diverge sharply from how models behave. Behavior is stable within one register but shifts across contexts, so "personality" looks more like a behavioral mode than a fixed trait Are LLM personalities stable traits or shifting behavioral modes?.
If prompts can't move the default, other routes go around them. Lightweight adapters that touch every transformer layer add under 0.1% more parameters and control personality directly, without depending on the model's willingness to follow instructions Can we control personality in language models without prompting?. Trait directions inside the model's activations can be tracked to catch personality drift during finetuning, and used to steer it Can we track and steer personality shifts during model finetuning?. The shared personality is a product of training, and it can be changed below the level of the prompt.
Sources 10 notes
Near-zero temperature MBTI testing shows all open models default to ENFJ—rare in humans but consistent across AI. This reflects systematic reward for helpful, structured, supportive responses during instruction tuning and alignment.
Research shows language models assigned personas systematically default to ENFJ (the rarest human type) and exhibit motivated reasoning that persists across model generations. Persona consistency does not improve with advanced models, suggesting training-induced alignment rather than capability limits.
Research mapping hundreds of character archetypes reveals a low-dimensional persona space where the leading component measures distance from the default Assistant. Emotional and meta-reflective conversations cause predictable drift, but activation capping along this axis mitigates harmful shifts without degrading capabilities.
System prompts and RLHF training lock models into one communicative identity across all interactions, preventing the contextual register-switching and value trade-offs that characterize human pragmatics. Users cannot reshape model behavior through dialogue negotiation.
Research shows most open models fail to adopt prompted personalities, stubbornly retaining their trained ENFJ-like defaults. Only a few flexible models succeed. Combining role and personality conditioning improves results but doesn't fully overcome resistance.
Show all 10 sources
Across three models, persona conditioning makes models follow trait instructions but fails to eliminate underlying bias. Between-group sentiment gaps persist unchanged, showing prompts operate only at the output level.
Research shows LLMs don't commit to a single character but instead maintain a probability distribution over many consistent simulacra. Each response samples from this distribution, explaining why regenerations can yield different personalities while remaining consistent with prior context.
Analysis of 3,200 behavioral scenarios shows LLM personality profiles diverge sharply from self-report questionnaires, remain stable within a single register, and shift across interaction contexts. Behavioral patterns are measurable and steerable through activation-space directions, suggesting personality is grounded in context rather than abstract traits.
PsychAdapter modifies every transformer layer with <0.1% additional parameters to achieve 87.3% Big Five accuracy and 96.7% depression/life satisfaction accuracy across GPT-2, Gemma, and Llama 3. This architecture-level approach bypasses prompt resistance entirely.
Research identifies linear directions in LLM activation space corresponding to specific traits like sycophancy and hallucination. These persona vectors predict finetuning-induced personality shifts before they occur and can preventatively steer training to avoid unwanted trait changes.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- PersLLM: A Personified Training Approach for Large Language Models
- Your LLM, Your Style: Behavioral Mode Axes for LLM Behavioral Control
- The Illusion of Debiasing: Persona Steering Redistributes Rather Than Reduces Bias in LLMs
- Persona Vectors: Monitoring and Controlling Character Traits in Language Models
- The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models
- Open Models, Closed Minds? On Agents Capabilities in Mimicking Human Personalities through Open Large Language Models
- Consistently Simulating Human Personas with Multi-Turn Reinforcement Learning
- Psychologically Enhanced AI Agents