analiza aiqbench
This research demonstrates that persona-based prompt engineering does not fundamentally alter the internal structure of large language models but merely modulates their output channel, often redistributing rather than eliminating bias. The study reveals a critical structural limit where surface-level steering fails to correct deep-seated disparities between groups while potentially amplifying specific closed-form biases. For developers and researchers, this underscores that relying on prompt interventions alone is insufficient for achieving genuine debiasing without addressing the model's underlying parametric associations.
The Illusion of Debiasing: Persona Steering Redistributes Rather Than Reduces Bias in LLMs
arXiv:2609.07117v1 Announce Type: new Abstract: Prompt-based interventions: system prompts, personas, role instructions, reliably reshape what a language model says, but it is unclear which layer they reach. Do they reconfigure internal structure, or only modulate the output channel? We use persona conditioning as a controlled probe, measuring its effects along a depth axis from self-report, through open-ended generation, to word-level parametric association, across three instruction-tuned models. We find a graded dissociation. Personas are legible but not structural: models follow single-trait instructions yet fail to reproduce human inter-trait covariance. The dissociation deepens with depth: personas hold or amplify closed-form QA bias, shift absolute tone while leaving between-group disparity unchanged, and barely perturb an already saturated associative baseline. Prompt-based steering thus operates in the output channel and has a structural reach limit that surface manipulability can mask.
Abstract:Prompt-based interventions: system prompts, personas, role instructions, reliably reshape what a language model says, but it is unclear which layer they reach. Do they reconfigure internal structure, or only modulate the output channel? We use persona conditioning as a controlled probe, measuring its effects along a depth axis from self-report, through open-ended generation, to word-level parametric association, across three instruction-tuned models. We find a graded dissociation. Personas are legible but not structural: models follow single-trait instructions yet fail to reproduce human inter-trait covariance. The dissociation deepens with depth: personas hold or amplify closed-form QA bias, shift absolute tone while leaving between-group disparity unchanged, and barely perturb an already saturated associative baseline. Prompt-based steering thus operates in the output channel and has a structural reach limit that surface manipulability can mask.
Computation and Language (cs.CL)
arXiv:2609.07117 [cs.CL]
(or arXiv:2609.07117v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.07117 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 225 cuvinte