What Is Mechanistic Interpretability? Inside AI Models
Mechanistic interpretability tries to read how AI models work inside. Features, circuits, superposition and sparse autoencoders, explained in plain English.

Mechanistic interpretability tries to reverse-engineer the circuits inside transformers. Posts explain induction heads, QK and OV circuits, feed-forward layers as key-value memories, sparse autoencoders, superposition and activation patching.
Mechanistic interpretability tries to read how AI models work inside. Features, circuits, superposition and sparse autoencoders, explained in plain English.

Activation Patching and Causal Tracing: Mathematical Foundations, Counterfactual Interventions, and Circuit Discovery in Large Language Models Mechanistic interpretability aims to reverse-engineer the computational graphs implemented by trained neural networks into human-understandable algorithms. In autoregressive Transformer models, standard observational tools such as attention map visualization and linear probing often fail to distinguish between active computational paths and passive corre

The mathematics of sparse autoencoders: superposition geometry, the L1 objective and its shrinkage bias, Top-K and JumpReLU sparsity, dead latents, evaluation metrics and feature steering.

The mathematics of activation addition and representation engineering: how concept directions are extracted, added or projected out, and how a residual stream shift reaches the output logits.

Induction Heads: Mathematical Foundations, Two-Layer Circuit Composition, and the Emergence of In-Context Learning in Transformers One of the defining capabilities of modern autoregressive large language models is in-context learning: the ability to infer rules, adapt to task formats, and execute complex few-shot instructions purely from prompt context without updating neural network weights. For years, in-context learning was treated as an enigmatic, emergent property of large-scale autoregres

The QK and OV Circuits in Transformers: How Bilinear Attention Routing and Subspace Projections Move Information The standard mathematical presentation of multi-head self-attention, introduced in Vaswani et al. (2017), describes the layer as a sequence of matrix projections followed by scaled dot-product operations, concatenation, and an output projection. While computationally efficient for parallel GPU hardware, this formulation obscures the fundamental linear mechanics governing how transfor

In transformer language models, multi-head self-attention and feed-forward networks (FFNs) perform distinct computational roles. While self-attention mechanisms route contextual information dynamically across sequence positions, feed-forward layers account for roughly two-thirds of total model parameters in standard architectures. Despite early interpretations characterizing FFNs simply as generic non-linear projections, mechanistic analysis shows that feed-forward layers operate as unnormalized

Technical deep dive on superposition in neural networks: almost-orthogonal vectors, sparsity, interference and why polysemantic neurons form in LLMs.

The Linear Representation Hypothesis in Large Language Models: How Neural Geometry Organizes Concepts, Latent Truth, and Steerable Features Deep neural networks are composed of alternating non-linear activation functions, multi-head attention operations, and high-dimensional matrix projections. Despite this architectural non-linearity, empirical research across mechanistic interpretability reveals a striking geometric regularity: within the intermediate representation spaces of large language m

Understanding how large language models perform complex reasoning requires moving beyond passive behavioral evaluation. While behavioral benchmarks measure model outputs on specific datasets, they treat the underlying neural network as an inscrutable black box. Mechanistic interpretability aims to reverse-engineer transformer weights and intermediate representations into human-understandable algorithms and computational graphs. At the core of modern mechanistic interpretability is causal mediat

Autoregressive large language models process input tokens through deep stacks of transformer blocks, generating a final probability distribution over the vocabulary at the final layer. For years, the intermediate computations occurring within these hidden layers were treated as opaque black boxes. Mechanistic interpretability research has demonstrated that modern transformer architectures operate through a structured, iterative refinement process across their residual stream. The logit lens and
