Superscopes: Amplifying Internal Feature Representations for Language Model Interpretation

TL;DR

This work addresses the challenge of interpreting internal representations in large language models by introducing Superscopes, an amplification-based extension to Patchscopes that magnifies weak, superposed features in $mlp$ outputs and hidden states before patching them into prompts. Grounded in the features-as-directions view and a CFG-inspired amplification concept, Superscopes reveals interpretable semantics without additional training and supports automatic amplifier detection. Through experiments across prompts and layers, amplified signals yield coherent interpretations and demonstrate the tool's flexibility for interpreting multiple internal components. The approach advances mechanistic interpretability and suggests future exploration of applying amplification to attention signals and refining amplifier selection, with a practical, open-source framework to facilitate broader use.

Abstract

Understanding and interpreting the internal representations of large language models (LLMs) remains an open challenge. Patchscopes introduced a method for probing internal activations by patching them into new prompts, prompting models to self-explain their hidden representations. We introduce Superscopes, a technique that systematically amplifies superposed features in MLP outputs (multilayer perceptron) and hidden states before patching them into new contexts. Inspired by the "features as directions" perspective and the Classifier-Free Guidance (CFG) approach from diffusion models, Superscopes amplifies weak but meaningful features, enabling the interpretation of internal representations that previous methods failed to explain-all without requiring additional training. This approach provides new insights into how LLMs build context and represent complex concepts, further advancing mechanistic interpretability.

Figures (8)

• Figure 1: An example of applying Patchscopes to the token “Live” from the prompt “Saturday Night Live” at layer 2, before hidden state contextualization occurs (first observed at layer 3). The raw MLP output appears nonsensical; however, once amplified, it correctly reflects the contextual meaning of the sentence.
• Figure 2: An example of Stable Diffusion with the text guidance "dog" and varying CFG guidance scales.
• Figure 2: Another example of MLP output interpretation. This example applies Patchscopes to the token “Future” from the prompt “Back to the Future” at layer 1, before hidden state contextualization occurs (first observed at layer 2). The raw MLP output relates to "Future" but lacks contextualized meaning. However, once amplified, it correctly reflects the contextual meaning of the sentence—referring to the 1985 science fiction movie Back to the Future, starring Marty McFly.
• Figure 3: A graph measuring the amount of successful interpretations of MLP outputs, by Superscopes and Patchscopes, over different layers. See §\ref{['subsec:amp_mlp_outputs']} for further analysis.
• Figure 3: An example of applying Patchscopes to the token “Great” from the prompt “Alexander the Great” at layers 4, yields "Barack Obama", while the Superscoped interpretation yields a correct interpretation ("Ancient Greek King"). Similarly, at layer 5, Patchscopes yields a "Barack Obama" while Superscopes yields an even more precise interpretation("of Macedon").