Skip to content
"Towards Monosemanticity: Decomposing Language Models With Dictionary Learning" by Zac Hatfield-Dodds episode artwork

Oct 9, 2023 4 min

"Towards Monosemanticity: Decomposing Language Models With Dictionary Learning" by Zac Hatfield-Dodds

from LessWrong (Curated & Popular)

Neural networks are trained on data, not programmed to follow rules. We understand the math of the trained network exactly – each neuron in a neural network performs simple arithmetic – but we don't understand why those mathematical operations result in the behaviors we see. This makes it hard to diagnose failure modes, hard to know how to fix them, and hard to certify that a model is truly safe. Luckily for those of us trying to understand artificial neural networks, we can simultaneously record the activation of every neuron in the network, intervene by silencing or stimulating them, and test the network's response to any possible input. This is a linkpost for https://transformer-circuits.pub/2023/monosemantic-features/ Text of post based on our blog post as a linkpost for the full paper which is considerably longer and more detailed. Source: https://www.lesswrong.com/posts/TDqvQFks6TWutJEKu/towards-monosemanticity-decomposing-language-models-with Narrated for LessWrong by TYPE III AUDIO. Share feedback on this narration. [125+ Karma Post] ✓

Episode details come from the podcast's RSS feed.

Embed this episode

Neural networks are trained on data, not programmed to follow rules. We understand the math of the trained network exactly – each neuron in a neural network performs simple arithmetic – but we don't understand why those mathematical operations result in the behaviors we see. This makes it hard to diagnose failure modes, hard to know how to fix them, and hard to certify that a model is truly safe. Luckily for those of us trying to understand artificial neural networks, we can simultaneously r...

Ready to play

"Towards Monosemanticity: Decomposing Language Models With Dictionary Learning" by Zac Hatfield-Dodds

0:00 4:55

No transcript yet.

Requests help us decide what to transcribe next.

Frequently asked questions

How long is this episode of LessWrong (Curated & Popular)?

This episode is 4 minutes long.

When was this LessWrong (Curated & Popular) episode published?

This episode was published on October 9, 2023.

Can I download this LessWrong (Curated & Popular) episode?

Yes. Use the download control on the episode player to save the publisher-provided media file.