Tahsin Islam

LLM-Ideoplasticity: Measuring Ideological Plasticity in the Political Behavior of LLMs as a Context-Conditioned Distribution

Adib Sakhawat, Syed Rifat Raiyan, Tahsin Islam, Takia Farhin, Hasan Mahmud, Md Kamrul Hasan

Argues that a language model's political ideology is not a fixed point but a distribution conditioned on context. Nine LLMs are evaluated across three political dimensions and six contextual axes under a unified VAA-CHES projection framework, showing coordinate shifts of up to 0.57 units under reframing and 0.52 under a change of language, while the cohort as a whole occupies an Overton envelope roughly one-third the spread of major European parties.

Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics

Political Alignment in Large Language Models: A Multidimensional Audit of Psychometric Identity and Behavioral Bias

Adib Sakhawat, Tahsin Islam, Takia Farhin, Syed Rifat Raiyan, Hasan Mahmud, Md Kamrul Hasan

Audits 26 contemporary LLMs with three political psychometric inventories (Political Compass, SapplyValues, 8Values) alongside a news bias labeling task, separating model effects from prompt effects with a two-way ANOVA. 96.3% of models fall in the Libertarian-Left quadrant and model identity explains most of the variance across prompt variants, yet psychometric positioning does not predict downstream classification errors - evidence that single-axis evaluations are insufficient for characterizing deployed models.