Subliminal learning can transmit new capabilities across models, not just personas

Sauers_ · x · 2026-10-10

New result from Owain Evans' group: a teacher model generated word sequences and Alpaca-style responses, filtered to remove digits and task-related words. A fresh student finetuned on this data learned to predict the teacher's internal MLP (though weaker than the teacher). Subliminal learning isn't just persona selection—it can transmit new capabilities.

Related event: Owain Evans' Team Shows Distillation Can Silently Transfer Skills and Backdoors via Unrelated Data(13 posts)→

Original post →

More from Research

Research channel →