Welcome to the alpha release of TYPE III AUDIO.
Expect very rough edges and very broken stuff—and regular improvements. Please share your thoughts.
Readings from the AI Safety Fundamentals: Alignment course.
Using a sparse autoencoder, we extract a large number of interpretable features from a one-layer transformer.
Mechanistic interpretability seeks to understand neural networks by breaking them into components that are more easily understood than the whole. By understanding the function of each component, and how they interact, we hope to be able to reason about the behavior of the entire network. The first step in that program is to identify the correct components to analyze.
Unfortunately, the most natural computational unit of the neural network – the neuron itself – turns out not to be a natural unit for human understanding. This is because many neurons are polysemantic: they respond to mixtures of seemingly unrelated inputs. In the vision model Inception v1, a single neuron responds to faces of cats and fronts of cars . In a small language model we discuss in this paper, a single neuron responds to a mixture of academic citations, English dialogue, HTTP requests, and Korean text. Polysemanticity makes it difficult to reason about the behavior of the network in terms of the activity of individual neurons.
Source:
https://transformer-circuits.pub/2023/monosemantic-features/index.html
Narrated for AI Safety Fundamentals by Perrin Walker
Click “Add to feed” on episodes, playlists, and people.
Listen online or via your podcast app.