Vision training doesn't universally improve AI language model human alignment

Original: VLMs May Not Globally Enhance Human Alignment over LLMs During Natural Reading

Source: arxiv.org ↗