SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2604.11048 · 2026

A Systematic Analysis of the Impact of Persona Steering on LLM Capabilities

Shi Feng, Jiaqi Chen, Ming Wang, Yongkang Liu, Tingna Xie

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 22 functions out of this paper's own repositories and ran 22 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
cjia7/DPR canonical 22 of 22
FunctionStatusWhere it lives
build_messages Ran cjia7/DPR/src/npti/neuron/apply_neuron_steering.py
pointer only (licence: NONE) · get_code("d3120aa76158afc9")
build_prompt Ran cjia7/DPR/src/npti/eval/gpt4_score.py
pointer only (licence: NONE) · get_code("16529904d2d5be9f")
calculate_quantiles Ran cjia7/DPR/src/npti/neuron/process_neuron.py
pointer only (licence: NONE) · get_code("75aa9b42ee1374ef")
extract_again Ran cjia7/DPR/src/npti/eval/eval_mmlu.py
pointer only (licence: NONE) · get_code("2e2c1555659e043f")
extract_answer Ran cjia7/DPR/src/npti/eval/eval_mmlu.py
pointer only (licence: NONE) · get_code("0ad16da4e4f52ed7")
extract_final Ran cjia7/DPR/src/npti/eval/eval_mmlu.py
pointer only (licence: NONE) · get_code("55653ca966a845f6")
extract_number Ran cjia7/DPR/src/npti/eval/eval_gsm8k.py
pointer only (licence: NONE) · get_code("59c6ffbd98640ae2")
extract_predicted_answer Ran cjia7/DPR/src/npti/eval/eval_bbh.py
pointer only (licence: NONE) · get_code("929072a79cd1c73f")
extract_predicted_index Ran cjia7/DPR/src/npti/eval/eval_musr.py
pointer only (licence: NONE) · get_code("1a763a2dbd070c96")
last_boxed_only_string Ran cjia7/DPR/src/npti/eval/eval_math.py
pointer only (licence: NONE) · get_code("0975c53e2a5fc2ab")
load_jsonl Ran cjia7/DPR/src/npti/eval/eval_bbh.py
pointer only (licence: NONE) · get_code("8fe69f136a96c214")
load_jsonl Ran cjia7/DPR/src/npti/eval/gpt4_score.py
pointer only (licence: NONE) · get_code("c7a248507261f132")
load_personality_descriptions Ran cjia7/DPR/src/npti/neuron/search_neuron.py
pointer only (licence: NONE) · get_code("d9495551b5d8cd4f")
load_questions Ran cjia7/DPR/src/npti/neuron/apply_neuron_steering.py
pointer only (licence: NONE) · get_code("e2da9069e83fab5e")
load_sorted_tuples Ran cjia7/DPR/src/npti/neuron/process_neuron.py
pointer only (licence: NONE) · get_code("2e02b59ac05c3c0f")
make_mlp_forward Ran cjia7/DPR/src/npti/neuron/search_neuron.py
pointer only (licence: NONE) · get_code("5594b19b20f7e8b2")
normalize_answer Ran cjia7/DPR/src/npti/eval/eval_bbh.py
pointer only (licence: NONE) · get_code("caa05d10c796877f")
normalize_final_answer Ran cjia7/DPR/src/npti/eval/eval_math.py
pointer only (licence: NONE) · get_code("4f5267c667058423")
parse_sampled_answer Ran cjia7/DPR/src/npti/eval/eval_gpqa.py
pointer only (licence: NONE) · get_code("91da82f1ebb2518b")
parse_strengths Ran cjia7/DPR/src/npti/neuron/apply_neuron_steering.py
pointer only (licence: NONE) · get_code("cbbf00812e6992fd")
process_differences Ran cjia7/DPR/src/npti/neuron/process_neuron.py
pointer only (licence: NONE) · get_code("dbd00e000f9c65dd")
remove_boxed Ran cjia7/DPR/src/npti/eval/eval_math.py
pointer only (licence: NONE) · get_code("ed70d70cd67847d0")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Imbuing Large Language Models (LLMs) with specific personas is prevalent for tailoring interaction styles, yet the impact on underlying cognitive capabilities remains unexplored. We employ the Neuron-based Personality Trait Induction (NPTI) framework to induce Big Five personality traits in LLMs and evaluate performance across six cognitive benchmarks. Our findings reveal that persona induction produces stable, reproducible shifts in cognitive task performance beyond surfacelevel stylistic changes. These effects exhibit strong task dependence: certain personalities yield consistent gains on instruction-following, while others impair complex reasoning. Effect magnitude varies systematically by trait dimension, with Openness and Extraversion exerting the most robust influence. Furthermore, LLM effects show 73.68% directional consistency with human personality-cognition relationships. Capitalizing on these regularities, we propose Dynamic Persona Routing (DPR), a lightweight query-adaptive strategy that outperforms the best static persona without additional training. Code and data are available at: https://github.com/cjia7/DPR.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2604.11048")
get_code_for_paper("2604.11048")
have("2604.11048")

Connect an agent — have() is free.