Qwen researcher: human annotation may be noisier than auto-labeling

antoine_chaffin · x · 2026-09-20

Antoine Chaffin, a multimodal researcher on the Qwen team, argues that recent "human annotated dataset" papers reveal human labels are often noisier and more error-prone than advanced automatic labeling. What works, he says, is iteration: let agents iterate repeatedly with human-readable summaries, potentially paired with human annotators.

Related event: Qwen researcher says human-labeled data may be noisier than auto-labeling(2 posts)→

Original post →

More from Research

Research channel →