Mechanistic Interpretability of Cognitive Complexity in LLMs via Linear Probing using Bloom's Taxonomy

TL;DR

This paper explores how large language models internally encode different levels of cognitive complexity, using Bloom's Taxonomy as a hierarchical framework, and demonstrates that these levels are linearly separable within the models' representations.

Contribution

It introduces a novel interpretability approach that uses linear probing with Bloom's Taxonomy to analyze cognitive complexity in LLMs, revealing hierarchical encoding of cognitive levels.

Findings

Linear classifiers achieve ~95% accuracy in identifying Bloom levels.

Cognitive levels are encoded in linearly separable subspaces.

Representations become more separable across layers.

Abstract

The black-box nature of Large Language Models necessitates novel evaluation frameworks that transcend surface-level performance metrics. This study investigates the internal neural representations of cognitive complexity using Bloom's Taxonomy as a hierarchical lens. By analyzing high-dimensional activation vectors from different LLMs, we probe whether different cognitive levels, ranging from basic recall (Remember) to abstract synthesis (Create), are linearly separable within the model's residual streams. Our results demonstrate that linear classifiers achieve approximately 95% mean accuracy across all Bloom levels, providing strong evidence that cognitive level is encoded in a linearly accessible subspace of the model's representations. These findings provide evidence that the model resolves the cognitive difficulty of a prompt early in the forward pass, with representations becoming increasingly separable across layers.