Tag
This paper formally defines the problem of eliciting latent knowledge (ELK) from AI systems using Causal Influence Diagrams, and proves an impossibility theorem: no feedback-based training strategy that depends only on agent behavior can guarantee an honest agent, even with perfect training feedback.
This paper argues that catastrophic forgetting in neural networks is not erasure but an interface alignment problem. It introduces 'transport keys' to recover latent task-specific features from sequentially trained models, demonstrating significant performance recovery on split CIFAR-100.
MechELK is a three-stage framework combining mechanistic interpretability tools (SAE, activation patching, causal probing) with representation engineering to elicit latent knowledge from LLMs, achieving 84.7% accuracy and outperforming existing methods like CCS and linear probing.