Encoded Early, Used Late: Where Transformers Begin to Act on an Inferred Partner's Expertise

Researchers found that transformers can encode an inferred partner's expertise in their residual stream at an early layer, before that information influences the model's output, extending prior findings on directly stated attributes to inferred attributes.

A transformer can make an attribute linearly decodable in its residual stream at a depth where that attribute does not yet influence the output. This gap between where information is readable and where it is used has been shown for attributes stated directly in the input. We ask whether it also hold