New research posted on arXiv takes a lifecycle view of occupational gender bias in large language models. The authors note that post-training is often assumed to reduce social biases that models pick up during pretraining, but they say direct evidence for that assumption is mixed. To address the gap, they examine how gender disparities evolve across the training pipeline.

Because the abstract is brief, the specific methods and results are not yet available in this summary. The paper's contribution, at this stage, is to frame the question empirically: rather than taking bias reduction for granted, it treats post-training as one phase whose effects need to be measured.

The preprint is listed as arXiv:2610.23587v1. As with any preprint, the work has not yet been peer-reviewed, and the abstract alone does not provide enough detail to assess the strength of the evidence.