))
Towards Monosemanticity: Decomposing Language Models With Dictionary Learning
Manage episode 409771548 series 3498845
Treść dostarczona przez BlueDot Impact. Cała zawartość podcastów, w tym odcinki, grafika i opisy podcastów, jest przesyłana i udostępniana bezpośrednio przez BlueDot Impact lub jego partnera na platformie podcastów. Jeśli uważasz, że ktoś wykorzystuje Twoje dzieło chronione prawem autorskim bez Twojej zgody, możesz postępować zgodnie z procedurą opisaną tutaj https://pl.player.fm/legal.
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
A podcast by BlueDot Impact.
Learn more on the AI Safety Fundamentals website.
64 odcinków
Udostępnij
