Sign inSign in
ConceptAI & ML

Mechanistic interpretability

Mechanistic interpretability tries to understand how individual components and operations in a machine learning model contribute to its overall decision-making process.
as Steven Feng defines itStanford UniversityThe wording closest to all 4 definitions
Watch it explained

Know first

Named as a prerequisite by the lectures that define this. The number is how many say so. A grey one has no page yet.

Neural networks3Large language models2Machine learning models1Network parameters1Neural network representations1

Lecturers

3

Each in their own terms, from the easiest level.

  • Steven Feng

    Stanford University · Undergraduate

    Techniques used to understand a model's internal representations, circuits, and features rather than just looking at its final outputs.
    • 1:08:43Overview of Transformers (Stanford CS25: Transformers United V6)
  • Ashish Vaswani, Lucas Beyer & William Fedus

    Stanford University · Graduate

    Mechanistic interpretability aims to map neural network parameters to human understandable algorithms.
    • 0:30Transformer Circuits, Induction Heads, In-Context Learning
  • Blake Richards

    Princeton University · Graduate

    The field of neuroscience and machine learning that studies how large language models solve problems by analyzing internal states and components like attention heads.
    • 1:11:03Revisiting the Turing Test in the Age of Large Language Models

Connections

6

The lectures that define it or assume it, and the concepts next to it.

3 lecturers. Definitions are written by AI from each lecture’s transcript. A time with a play mark starts the lecture at the moment the lecturer says it.