Tag
This article presents a Tarski-inspired diagonal argument showing that no linear probe on an LLM's embedding space can reliably detect truth, drawing parallels to Gödel's incompleteness and Turing's halting problem. It critiques the linear representation hypothesis for truth in language models.
This paper investigates the geometric structure of intermediate feature representations in deep neural networks by analyzing how various image manipulations map in feature space. It suggests that feature spaces are organized in linear structures to a first approximation, using generative image editing models to probe these representations.