A new preprint on arXiv proposes using large language models for personality assessment in asynchronous video interviews (AVIs), a method increasingly used in recruitment to standardize evaluation. The authors position their work as addressing a gap in prior research, which the abstract describes as confined in some unspecified way.
The stated goal is to align behavioral evidence extracted by LLMs with human ratings of personality. This suggests an attempt to make automated scoring more consistent with human judgment, potentially improving the validity of AVI-based assessments.
However, the available abstract is truncated, so details on the model architecture, dataset, and evaluation results are not included. The claims should therefore be read as preliminary, pending the full paper.