Read the original at HF Daily Papers
Researchers investigate the Platonic Representation Hypothesis, finding that cross-modal alignment persists for newer models with clean captions and can exceed explicitly aligned vision-language models.
Carried by: HF Daily Papers. First seen: .