cs.CLApr 13, 2026

Psychological Concept Neurons: Can Neural Control Bias Probing and Shift Generation in LLMs?

Authors: Yuto Harada, Hiro Taiyo Hamada

Organizations: Araya Inc.

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Using psychological constructs such as the Big Five, large language models (LLMs) can imitate specific personality profiles and predict a user's personality. While LLMs can exhibit behaviors consistent with these constructs, it remains unclear where and how they are represented inside the model and how they relate to behavioral outputs. To address this gap, we focus on questionnaire-operationalized Big Five concepts, analyze the formation and localization of their internal representations, and use interventions to examine how these representations relate to behavioral outputs. In our experiment, we first use probing to examine where Big Five information emerges across model depth. We then identify neurons that respond selectively to each Big Five concept and test whether enhancing or suppressing their activations can bias latent representations and label generation in intended directions. We find that Big Five information becomes rapidly decodable in early layers and remains detectable through the final layers, while concept-selective neurons are most prevalent in mid layers and exhibit limited overlap across domains. Interventions on these neurons consistently shift probe readouts toward targeted concepts, with targeted success rates exceeding 0.8 for some concepts, indicating that the model's internal separation of Big Five personality traits can be causally steered. At the label-generation level, the same interventions often bias generated label distributions in the intended directions, but the effects are weaker, more concept-dependent, and often accompanied by cross-trait spillover, indicating that comparable control over generated labels is difficult even with interventions on a large fraction of concept-selective neurons. Overall, our findings reveal a gap between representational control and behavioral control in LLMs.

Figures & tables

Figure 1: Overview of the background and experimental design. (a) The Big Five framework defines human personality using five latent trait domains, which are treated in psychology as psychological constructs (i.e., abstract properties that are not directly observable but operationally defined). (b) Example item from the Big Five Inventory-2 (BFI-2), a standard personality questionnaire, for Extraversion. Respondents rate how well the statement describes them on a 5-point Likert scale. (c) Related trends. Recent work increasingly applies psychometric personality concepts to LLMs, showing that models can exhibit and be conditioned toward stable personality-like behaviors. However, these findings are largely behavioral, leaving the underlying internal representations and mechanisms unclear. (d) Our approach is organized around three questions: RQ1 where questionnaire-defined concepts emerge across layers, RQ2 which neurons are selective for each concept, and RQ3 whether intervening on those neurons can causally steer readouts and label generation.
Figure 2: A two-dimensional visualisation of the residual stream used in the probing experiment. Each point corresponds to a questionnaire item, colored by its domain label, and the layer index is indicated in the panel title. S denotes silhouette score, D denotes intra-cluster mean distance.
Figure 3: Results of layer-wise probing experiments. The x-axis indicates the model layer, which varies between models; hence it is shown as a proportion from the initial layer to the final layer. The results for each domain within the Big Five, with “Overall” representing the average result across all domains.
Figure 4: Results of probing experiments for each comparable model size. The dashed line indicates smaller model sizes within the same series.
Figure 5: Layer-wise distribution of concept-selective neurons. For each domain, we select the top 1000 units across the entire model and plot their layer locations.
Figure 6: Results of intervention experiments on layer-wise probing. We intervened in concept-selective neurons in two ways: activation boosting involved patching to the top 99th percentile of activation values in sentences representing the target concept, while activation suppressing involved patching to the bottom 1st percentile of activation values in sentences not representing the target concept. The figure on the right shows both applied.
Figure 7
Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Figure 9: Layer-wise trends of UMAP-based metrics for Big Five domain labels. Left: silhouette score SS (higher indicates clearer separation between domains). Right: intra-cluster distance DD (lower indicates tighter within-domain compactness).
More in this topic

Personality Modeling in Language Models

Personality modeling in language models studies how traits are measured, represented, simulated, and controlled in model behavior.

See all 169 papers

Explore similar work

CardsList
Aug 8, 2025

Scaling Personality Control in LLMs with Big Five Scaler Prompts

We present Big5-Scaler, a prompt-based framework for conditioning large language models (LLMs) with controllable Big Five personality traits. By embedding numeric trait values into natural language prompts, our method enables fine-grained personality control without additional training. We evaluate Big5-Scaler across trait expression, dialogue generation, and human trait imitation tasks. Results show that it induces consistent and distinguishable personality traits across models, with performance varying by prompt type and scale. Our analysis highlights the effectiveness of concise prompts and lower trait intensities, providing a efficient approach for building personality-aware dialogue agents.

Gunhee Cho, Yun-Gyung Cheong

Sungkyunkwan University, South Korea

Persona-Conditioned GenerationControllable Text GenerationPersonality Modeling in Language ModelsLLM Prompting
AskPDF
Apr 15, 2026

Psychological Steering of Large Language Models

Large language models (LLMs) emulate a consistent human-like behavior that can be shaped through activation-level interventions. This paradigm is converging on additive residual-stream injections, which rely on injection-strength sweeps to approximate optimal intervention settings. However, existing methods restrict the search space and sweep in uncalibrated activation-space units, potentially missing optimal intervention conditions. Thus, we introduce a psychological steering framework that performs unbounded, fluency-constrained sweeps in semantically calibrated units. Our method derives and calibrates residual-stream injections using psychological artifacts, and we use the IPIP-NEO-120, which measures the OCEAN personality model, to compare six injection methods. We find that mean-difference (MD) injections outperform Personality Prompting (P2^2), an established baseline for OCEAN steering, in open-ended generation in 11 of 14 LLMs, with gains of 3.6% to 16.4%, overturning prior reports favoring prompting and positioning representation engineering as a new frontier in open-ended psychological steering. Further, we find that a hybrid of P2^2 and MD injections outperforms both methods in 13 of 14 LLMs, with gains over P2^2 ranging from 5.6% to 21.9% and from 3.3% to 26.7% over MD injections. Finally, we show that MD injections align with the Linear Representation Hypothesis and provide reliable, approximately linear control knobs for psychological steering. Nevertheless, they also induce OCEAN trait covariance patterns that depart from the Big Two model, suggesting a gap between learned representations and human psychology.

Leonardo Blas, Robin Jia, Emilio Ferrara

Thomas Lord Department of Computer Science, University of Southern California, Los Angeles, CA 90089

Language Model SteeringRepresentation EngineeringPersonality Modeling in Language Models
AskPDF
Apr 30, 2026

DPN-LE: Dual Personality Neuron Localization and Editing for Large Language Models

With the widespread adoption of large language models (LLMs), understanding their personality representation mechanisms has become critical. As a novel paradigm in Personality Editing, most existing methods employ neuron-editing to locate and modify LLM neurons, requiring changes to numerous neurons and leading to significant performance degradation. This raises a fundamental question: Are all modified neurons directly related to personality representation? In this work, we investigate and quantify this specificity through assessments of general capability impact and representation-level patterns. We find that: 1) Current methods can change personalities but reduce overall performance. 2) Neurons are multifunctional, connecting personality traits and general knowledge. 3) Opposing personality traits demonstrate distinctly mutually exclusive representation patterns. Motivated by these findings, we propose DPN-LE (Dual Personality Neuron Localization and Editing), which identifies personality-specific neurons by contrasting MLP activations between high-trait and low-trait samples. DPN-LE constructs layer-wise steering vectors and applies dual-criterion filtering based on Cohen's dd effect size and activation magnitude to isolate mutually exclusive neuron subsets. Sparse linear intervention on these neurons enables precise personality control at inference time. Using only 1,000 contrastive sample pairs per trait, DPN-LE intervenes on ∼\sim0.5% of neurons while achieving competitive personality control and substantially better capability preservation across reasoning tasks. Experiments on LLaMA-3-8B-Instruct and Qwen2.5-7B-Instruct demonstrate the effectiveness and generalizability of our approach.

Lifan Zheng, Xue Yang, Jiawei Chen +6

Southeast University · Shanghai Jiao Tong University · East China Normal University +6

LLM InterpretabilityPersonality Modeling in Language ModelsCausal Interventions in Language Models
AskPDF