cs.AIFeb 19, 2026

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

Authors: Bianca Raimondi, Maurizio Gabbrielli

Organizations: Department of Computer Science and Engineering, University of Bologna, Bologna, Italy

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

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.

Figures & tables

Figure 1: Overview of the experimental pipeline, from dataset construction and activation extraction to layer-wise linear probing.
ModelLdmd_{m}
Llama-3.1-8B-Instruct324096
Qwen3-4B-Instruct-2507362560
gemma-3-4b-it352560
DeepSeek-R1-Distill-Llama-8B324096
Table 1: Models used in our study.
Bloom Leveld0d_{0}…dmd_{m}
00.0047…0.0007
10.0055…0.0029
50.0050…0.0019
30.0015…0.0075
Table 2: Example of stored activations at specific layer i.
Figure 2: Layer-wise probe accuracy across all evaluated models.
Figure 3: Probe accuracy at layer 5 across Bloom levels for four models. All architectures exhibit consistently high performance, with minor variations reflecting the cognitive difficulty.
Figure 4: Representative confusion matrix of the linear probe for Llama-3.1-8B-Instruct at layer 5.
Figure 5: Figure 5: Layer-wise Euclidean distances between adjacent Bloom-level centroids for the Llama-3.1-8B-Instruct model. Distances are small in early layers and increase monotonically with depth, indicating progressive geometric disentanglement of cognitive levels. The CSO layer l⋆=5l^{\star}=5 marks the onset of rapid separation.
Figure 6: Confusion matrices for the control experiments: TF-IDF on the left, and Sentence Embeddings on the right.
Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 7: Layer-wise linear probe accuracy for each Bloom level.
Figure 8: Confusion matrices of the linear probe for four representative models at the CSO layer l⋆=5l^{\star}=5 : gemma-3-4b-it , DeepSeek-R1-Distill-Llama-8B , and Qwen3-4B-Instruct-2507 .
Figure 9: Layer-wise Euclidean distances between adjacent Bloom-level centroids for all evaluated models: gemma-3-4b-it , DeepSeek-R1-Distill-Llama-8B , and Qwen3-4B-Instruct-2507 .
More in this topic

LLM Interpretability

Large language model (LLM) interpretability studies how internal representations and computations produce outputs, and how they can be explained or controlled.

See all 778 papers

Explore similar work

CardsList

  1. A Neuropsychologically Grounded Evaluation of LLM Cognitive Abilities

    Mar 3, 2026Faiz Ghifari Haznitrama, Faeyza Rishad Ardi, Alice OhLLM Evaluation

    School of Computing, KAIST, Daejeon, South Korea

  2. Beyond Linear Concepts: Discovering and Aligning Non-Linear Concept Manifolds in Large Language Models

    Oct 1, 2026Tido Specht, Elias Benedict Krey, Nils Neukirch +1LLM AlignmentLLM Interpretability

    University of Groningen · Carl von Ossietzky University of Oldenburg

  3. Modeling Multi-Dimensional Cognitive States in Large Language Models under Cognitive Crowding

    Apr 19, 2026Lin Zhong, Siyu Zhu, Zizhen Yuan +5LLM AlignmentCognitive Modeling

    Harbin Institute of Technology, Shenzhen, China · City University of Macau, Macao SAR, China · Peng Cheng Laboratory, Shenzhen, China