SYNTHESIS NOTE
Can open language models adopt different personalities through prompting?
Explores whether open LLMs can be conditioned to mimic target personalities via prompting, or whether they resist and retain their default traits regardless of instructions.
The "Open Models, Closed Minds" study tested whether open LLMs can mimic human personalities when conditioned through prompting. The finding: most cannot. When given personality-conditioning prompts, the majority of models retain their intrinsic traits — the ENFJ-like default — rather than shifting to the target personality. The authors call this being "closed-minded."
Only a few models (SOLAR, NeuralChat, Llama3-8, Dolphin) demonstrate genuine flexibility, successfully mirroring imposed personalities regardless of temperature setting. The rest are stubborn.
A partial solution emerges: combining role conditioning (e.g., "you are a dentist") with personality conditioning (e.g., "you are introverted and analytical") produces better results than personality conditioning alone. The ENFJ archetype — trained as a teacher — responds to being given a concrete professional role because roles provide behavioral anchors that abstract personality dimensions don't.
This is a different failure mode from Why do LLM persona prompts produce inconsistent outputs across runs?. That finding shows run-to-run instability — the model's output varies unpredictably under persona prompts. This finding shows resistance — the model's output remains stubbornly stable on its default personality regardless of prompts. Together they form two sides of a persona failure taxonomy:
- Instability: model generates varying outputs that reflect uncertainty, not persona knowledge
- Resistance: model retains intrinsic personality traits despite conditioning attempts
- Motivated reasoning: persona conditioning introduces cognitive biases (see Do personas make language models reason like biased humans?)
The practical implication: persona engineering requires more than prompting. Role-personality combinations work better than personality alone. But even then, model selection matters — most models simply cannot be steered to arbitrary personality configurations through in-context methods.
Inquiring lines that read this note 84
This note is a source for these research framings, grouped by the broader line of inquiry each explores. Scan the bold lines of inquiry; follow any specific question forward.
Can persona prompting reliably change how language models behave?- How do LLM user simulators fail to represent authentic user behavior distributions?
- Why do language models successfully simulate political perspectives and social personas?
- How do LLMs identify which personality items matter most for trait inference?
- Why do LLM regenerations produce meaningfully different personalities from the same prompt?
- How do LLM personas compare to demographic targeting?
- How does the dialogue prompt establish the character the model plays?
- Do personality traits occupy specific mechanistic locations in pretrained models?
- Do open-source LLMs show different resistance patterns to persona prompting than closed models?
- How does personality priming change LLM strategic decision making?
- Why do some open models resist personality conditioning while others don't?
- Can persona prompting overcome the default ENFJ personality in language models?
- Do training objectives directly determine the ENFJ default across models?
- How does RLHF-induced mode collapse limit diversity in LLM-generated personas?
- Do personality traits occupy consistent geometric structures across different LLM architectures?
- Why do language models resist adopting different personalities when prompted?
- What neural mechanisms in LLMs create or maintain simulated personality traits?
- Does pre-training encode personality patterns that fine-tuning later activates?
- Can we detect superposition in LLM personality traits and stated preferences?
- Why do LLM persona simulations replicate main effects but fail on marginal effects?
- Does richer input to LLM personas improve their fidelity to human responses?
- Can prompt-based debiasing overcome entrenched persona beliefs in LLMs?
- Why do questionnaire-based personality scores fail to predict actual LLM behavioral choices?
- What explains why LLM personas fail to instantiate values but succeed in sounding natural?
- How does behavioral stickiness distinguish realized from pretended personas?
- How do lightweight adapters modify model behavior for personality traits?
- Do personality traits and task knowledge occupy separate subspaces in transformer parameters?
- What specific character traits drive memory selection in persona-based retrieval?
- How do lightweight adapters control personality traits across different transformer layers?
- Can models transmit behavioral traits through semantically unrelated synthetic data?
- Can prompting inject new knowledge into already-trained AI models?
- Can prompting alone inject new domain knowledge into a model?
- Are instruction-tuned models more or less sensitive to prompt semantics than others?
- Can users inject entirely new knowledge into models through prompting alone?
- Can activation-space directions reliably steer LLM behavior without retraining or prompting?
- Can multimodal LLMs be made to spontaneously adapt their language for efficiency?
- Do different LLM model sizes or architectures show varying susceptibility to nudges?
- Can prompting a deceptive role change how an LLM tailors its lies?
- Can distinctive input voices maintain accuracy without adopting the model's preferred register?
- How does prompting language shift what LLMs express about political figures?
- Why do most open language models resist personality conditioning via prompts?
- How does prompt design alter what kind of creativity LLMs can express?
- Does combining role and personality prompts produce stable behavioral changes?
- Why do models resist personality change despite sophisticated prompting techniques?
- Why do language models prefer certain response styles regardless of what the prompt asks?
- How do LLM behavioral profiles differ across prompt registers like advice versus task execution?
- Do these characteristic phrases persist across newer model versions and different prompt types?
- What distinguishes character simulation from authentic voice in language model outputs?
- Why do different language models converge on similar narrative defaults?
- Do open language models default to a single shared personality type?
- How does language condition affect model psychological profile consistency?
- What role does prompt context play in preventing genuine addressee modeling in generation?
- Can role-play and explicit step-by-step prompting be isolated as separate mechanisms?
- Why can data filtering fail to remove transmitted behavioral traits?
- Can sycophancy be reliably steered through system prompts alone?
- What distinguishes personality resistance from persona instability in LLMs?
- Why does dynamic persona identification outperform fixed personas in prompting?
- Can dynamic personality modeling prevent the repetitiveness of static predefined personas?
- Why do personas in language models resist correction through prompting alone?
- Does linguistic style or content richness matter more for persona authenticity?
- How do personality and language proficiency moderate the impact of linguistic alignment?
- Can dynamic personality modeling without event-specificity produce plausible dialogue?
- How do language models transmit traits through semantically unrelated data?
- Can interventions on individual features reliably steer language model behavior?
- Can a model imitate ChatGPT's style by copying its phrases without matching its capabilities?
- Can style imitation training successfully erase a model's distinctive word preferences?
Related concepts in this collection 5
This note in its neighbourhood — explore the map, then jump to a related concept in the list below.
Concept map
Can open language models adopt different persona…
Why do LLM persona prompts produce inconsistent ou…
Why do open language models converge on one person…
Does model capability translate to better persona …
What anchors a stable identity beneath an LLM's pe…
How stable is the trained Assistant personality in…
Click a node to walk · click center to open · click Open in graph to see this note in the full knowledge graph
your link
semantically near
linked from elsewhere
-
Why do LLM persona prompts produce inconsistent outputs across runs?
Can language models reliably simulate different social perspectives through persona prompting, or does their run-to-run variance indicate they lack stable group-specific knowledge? This matters for whether LLMs can approximate human disagreement in annotation tasks.
complementary failure mode: instability vs resistance
-
Why do open language models converge on one personality type?
Research testing LLMs on personality metrics reveals consistent clustering around ENFJ—the rarest human type. This explores what training mechanisms drive this convergence and what it reveals about AI alignment.
the default personality that models resist changing
-
Does model capability translate to better persona consistency?
As language models become more advanced, do they naturally become better at maintaining consistent personas across conversations? PersonaGym testing across multiple models and thousands of interactions explores whether scaling helps with persona adherence.
capability scaling doesn't help either
-
What anchors a stable identity beneath an LLM's persona?
Human personas are grounded in biological needs and embodied experience, creating a stable self beneath social performance. Do LLMs have any comparable anchor, or is their identity purely situational?
personality resistance complicates the "nothing beneath" claim: the trained ENFJ default functions as a quasi-stable substrate that persists across prompting attempts, even though it's a training artifact rather than a biological self
-
How stable is the trained Assistant personality in language models?
Explores whether post-training successfully anchors models to their default Assistant mode, or whether conversations can predictably pull them toward different personas. Understanding persona stability matters for safety and reliability.
provides the geometric explanation for closed-mindedness: prompt-based personality conditioning may fail because it cannot shift activations far enough from the Assistant region of persona space; the "loose tethering" is what makes models resistant to prompt-level persona change
Related papers in this collection 8
Papers most semantically related to this note, ranked by cosine similarity in the embedding space.
- Open Models, Closed Minds? On Agents Capabilities in Mimicking Human Personalities through Open Large Language Models
- Psychologically Enhanced AI Agents
- Do LLMs Possess a Personality? Making the MBTI Test an Amazing Evaluation 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
- PersLLM: A Personified Training Approach for Large Language Models
- When "A Helpful Assistant" Is Not Really Helpful: Personas in System Prompts Do Not Improve Performances of Large Language Models
- Consistently Simulating Human Personas with Multi-Turn Reinforcement Learning
Original note title
most open LLMs are closed-minded to personality conditioning — retaining intrinsic traits despite prompting while combining role and personality conditioning partially overcomes resistance