Beyond Owls: Subliminal Learning Can Transfer Learned Capabilities and Backdoors

Key points

• Can SL transfer a wider range of traits, including more complex ones?1
• To this end, we test whether SL can transfer a novel capability: predicting the outputs of a randomly initialized MLP.1
• We find that a directly optimized steering vector matches SL in distribution but generalizes worse out of distribution.1
• Thus, we show SL can transfer capabilities, backdoors, and hacking propensities.1

Sources (1)

• [1]Beyond Owls: Subliminal Learning Can Transfer Learned Capabilities and Backdoors
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 04:42 PM

  In subliminal learning (SL), a teacher model passes on a trait to a student model by distillation on data semantically unrelated to the trait.

  Can SL transfer a wider range of traits, including more complex ones?

Extractive summary: sentences quoted from the sources.