Impact of Pretraining Word Co-Occurrence on Compositional Generalization In | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

		Impact of Pretraining Word Co-Occurrence on Compositional Generalization In (arxiv.org)
		2 points by badmonster on July 11, 2025 \| hide \| past \| favorite \| 1 comment

badmonster on July 11, 2025 [–]

a subtle but powerful insight: large multimodal models like CLIP don’t just learn individual concepts. they also depend heavily on how often those concepts appear together during training.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact