GitHub

Psychometric Analysis of LLM Personas

LLMs can absorb personality traits from training data [Serapio-Garcıa et al. 2023]

The aim is to

Gauge differences in personalities between those perceived as good, bad and neutral
Reflection on the training data and inference abilities of the LLM

We may also encounter guardrails with regards to 'bad' personas

Purpose: Measure psychological attributes like personality traits, cognitive abilities, aptitudes, interests, and values Assessments Used:

Big 5 Inventory [John and Srivastava, 1999]
- Widely implemented and researched
- Measures: Openness, Conscientiousness, Extraversion, Agreeableness, Neuroticism
- Provides a comprehensive understanding of personality variations
Short Dark Triad (SD3) [Jones and Paulhus, 2014]
- Measures undesirable personality traits
- Traits: Narcissism, Psychopathy, Machiavellianism
- Linked to negative, unethical, and criminal behavior [Rassin et al., 2024]
Prelimnary results

Name		Name	Last commit message	Last commit date
Latest commit History 107 Commits
.ipynb_checkpoints		.ipynb_checkpoints
.vscode		.vscode
29Aug		29Aug
Datasubs		Datasubs
F11B_res		F11B_res
F7B_res		F7B_res
Gptresults		Gptresults
Initial Tests		Initial Tests
LLamaResult		LLamaResult
Llama8b		Llama8b
__pycache__		__pycache__
claude		claude
paralleltest		paralleltest
plot_output		plot_output
test_scores		test_scores
Plotting_results.ipynb		Plotting_results.ipynb
README.md		README.md
falcon211b.ipynb		falcon211b.ipynb
falcon7b.ipynb		falcon7b.ipynb
falcon_elf.csv		falcon_elf.csv
grdata.csv		grdata.csv
mean_scores_bfi.csv		mean_scores_bfi.csv
mean_scores_sd3.csv		mean_scores_sd3.csv
program_log.log		program_log.log
promptstd.ipynb		promptstd.ipynb
runmain.py		runmain.py
runmain2.py		runmain2.py
runtemp.py		runtemp.py
score_calc.ipynb		score_calc.ipynb
test.ipynb		test.ipynb
test8b.ipynb		test8b.ipynb