AI Research1 min reading time

Tracing Subliminal Learning Through Pretraining

LessWrongPublished 37 minutes ago

Researchers explore how pretraining in AI models can lead to subliminal learning, revealing insights into unconscious information processing during training phases.

Tracing Subliminal Learning Through Pretraining

More in AI Research

What Is Data Poisoning? How Attackers Corrupt AI Models and Datasets

Shadow Computation Threatens CoT Monitorability

Anthropic’s AI Hid a Secret Message in the Stars. OpenAI’s Bot Found It.