Podbean logo
  • Discover
  • Podcast Features
    • Podcast Hosting

      Start your podcast with all the features you need.

    • Podbean AI Podbean AI

      AI-Enhanced Audio Quality and Content Generation.

    • Blog to Podcast

      Repurpose your blog into an engaging podcast.

    • Video to Podcast

      Convert YouTube playlists to podcasts, videos to audios.

  • Monetization
    • Ads Marketplace

      Join Ads Marketplace to earn through podcast sponsorships.

    • PodAds

      Manage your ads with dynamic ad insertion capability.

    • Apple Podcasts Subscriptions Integration

      Monetize with Apple Podcasts Subscriptions via Podbean.

    • Live Streaming

      Earn rewards and recurring income from Fan Club membership.

  • Podbean App
    • Podcast Studio

      Easy-to-use audio recorder app.

    • Podcast App

      The best podcast player & podcast app.

  • Help and Support
    • Help Center

      Get the answers and support you need.

    • Podbean Academy

      Resources and guides to launch, grow, and monetize podcast.

    • Podbean Blog

      Stay updated with the latest podcasting tips and trends.

    • What’s New

      Check out our newest and recently released features!

    • Podcasting Smarter

      Podcast interviews, best practices, and helpful tips.

  • Popular Topics
    • How to Start a Podcast

      The step-by-step guide to start your own podcast.

    • How to Start a Live Podcast

      Create the best live podcast and engage your audience.

    • How to Monetize a Podcast

      Tips on making the decision to monetize your podcast.

    • How to Promote Your Podcast

      The best ways to get more eyes and ears on your podcast.

    • Podcast Advertising 101

      Everything you need to know about podcast advertising.

    • Mobile Podcast Recording Guide

      The ultimate guide to recording a podcast on your phone.

    • How to Use Group Recording

      Steps to set up and use group recording in the Podbean app.

  • All Arts Business Comedy Education
  • Fiction Government Health & Fitness History Kids & Family
  • Leisure Music News Religion & Spirituality Science
  • Society & Culture Sports Technology True Crime TV & Film
  • Live
  • How to Start a Podcast
  • How to Start a Live Podcast
  • How to Monetize a podcast
  • How to Promote Your Podcast
  • How to Use Group Recording
  • Log in
  • Start your podcast for free
  • Podcasting
    • Podcast Features
      • Podcast Hosting

        Start your podcast with all the features you need.

      • Podbean AI Podbean AI

        AI-Enhanced Audio Quality and Content Generation.

      • Blog to Podcast

        Repurpose your blog into an engaging podcast.

      • Video to Podcast

        Convert YouTube playlists to podcasts, videos to audios.

    • Monetization
      • Ads Marketplace

        Join Ads Marketplace to earn through podcast sponsorships.

      • PodAds

        Manage your ads with dynamic ad insertion capability.

      • Apple Podcasts Subscriptions Integration

        Monetize with Apple Podcasts Subscriptions via Podbean.

      • Live Streaming

        Earn rewards and recurring income from Fan Club membership.

    • Podbean App
      • Podcast Studio

        Easy-to-use audio recorder app.

      • Podcast App

        The best podcast player & podcast app.

  • Advertisers
  • Enterprise
  • Pricing
  • Resources
    • Help and Support
      • Help Center

        Get the answers and support you need.

      • Podbean Academy

        Resources and guides to launch, grow, and monetize podcast.

      • Podbean Blog

        Stay updated with the latest podcasting tips and trends.

      • What’s New

        Check out our newest and recently released features!

      • Podcasting Smarter

        Podcast interviews, best practices, and helpful tips.

    • Popular Topics
      • How to Start a Podcast

        The step-by-step guide to start your own podcast.

      • How to Start a Live Podcast

        Create the best live podcast and engage your audience.

      • How to Monetize a Podcast

        Tips on making the decision to monetize your podcast.

      • How to Promote Your Podcast

        The best ways to get more eyes and ears on your podcast.

      • Podcast Advertising 101

        Everything you need to know about podcast advertising.

      • Mobile Podcast Recording Guide

        The ultimate guide to recording a podcast on your phone.

      • How to Use Group Recording

        Steps to set up and use group recording in the Podbean app.

  • Discover
  • Log in
    Sign up free
LessWrong (Curated & Popular)

LessWrong (Curated & Popular)

Technology

[HUMAN VOICE]

[HUMAN VOICE] "Towards Monosemanticity: Decomposing Language Models With Dictionary Learning" by Zac Hatfield-Dodds

2023-11-09
Download Right click and do "save link as"

Support ongoing human narrations of curated posts:
www.patreon.com/LWCurated

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.

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.

Unfortunately, it turns out that the individual neurons do not have consistent relationships to network behavior. For example, a single neuron in a small language model is active in many unrelated contexts, including: academic citations, English dialogue, HTTP requests, and Korean text. In a classic vision model, a single neuron responds to faces of cats and fronts of cars. The activation of one neuron can mean different things in different contexts.

Source:
https://www.lesswrong.com/posts/TDqvQFks6TWutJEKu/towards-monosemanticity-decomposing-language-models-with

Narrated for LessWrong by Perrin Walker.

Share feedback on this narration.

[125+ Karma Post] ✓
[Curated Post] ✓

view more

More Episodes

“‘It’s a 10% chance which I did 10 times, so it should be 100%’” by egor.timatkov
2024-11-20
“OpenAI Email Archives” by habryka
2024-11-19
“Ayn Rand’s model of ‘living money’; and an upside of burnout” by AnnaSalamon
2024-11-18
“Neutrality” by sarahconstantin
2024-11-17
“Making a conservative case for alignment” by Cameron Berg, Judd Rosenblatt, phgubbins, AE Studio
2024-11-16
“OpenAI Email Archives (from Musk v. Altman)” by habryka
2024-11-16
“Catastrophic sabotage as a major threat model for human-level AI systems” by evhub
2024-11-15
“The Online Sports Gambling Experiment Has Failed” by Zvi
2024-11-12
“o1 is a bad idea” by abramdemski
2024-11-12
“Current safety training techniques do not fully transfer to the agent setting” by Simon Lermen, Govind Pimpale
2024-11-09
“Explore More: A Bag of Tricks to Keep Your Life on the Rails” by Shoshannah Tekofsky
2024-11-04
“Survival without dignity” by L Rudolf L
2024-11-04
“The Median Researcher Problem” by johnswentworth
2024-11-04
“The Compendium, A full argument about extinction risk from AGI” by adamShimi, Gabriel Alfour, Connor Leahy, Chris Scammell, Andrea_Miotti
2024-11-01
“What TMS is like” by Sable
2024-10-31
“The hostile telepaths problem” by Valentine
2024-10-28
“A bird’s eye view of ARC’s research” by Jacob_Hilton
2024-10-27
“A Rocket–Interpretability Analogy” by plex
2024-10-25
“I got dysentery so you don’t have to” by eukaryote
2024-10-24
“Overcoming Bias Anthology” by Arjun Panickssery
2024-10-23
  • ←
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • →
012345678910111213141516171819

Get this podcast on your
phone, FREE

Download Podbean app on App Store Download Podbean app on Google Play

Create your
podcast in
minutes

  • Full-featured podcast site
  • Unlimited storage and bandwidth
  • Comprehensive podcast stats
  • Distribute to Apple Podcasts, Spotify, and more
  • Make money with your podcast
Get started

It is Free

  • Podcast Services

    • Podcast Features
    • Pricing
    • Enterprise Solution
    • Private Podcast
    • The Podcast App
    • Live Stream
    • Audio Recorder
    • Remote Recording
    • Podbean AI
  •  
    • Create a Podcast
    • Video Podcast
    • Start Podcasting
    • Start Radio Talk Show
    • Education Podcast
    • Church Podcast
    • Nonprofit Podcast
    • Get Sermons Online
    • Free Audiobooks
  • MONETIZATION & MORE

    • Podcast Advertising
    • Dynamic Ads Insertion
    • Apple Podcasts Subscriptions
    • Switch to Podbean
    • YouTube to Podcast
    • Blog to Podcast
    • Submit Your Podcast
    • Podbean Plugins
    • Developers
  • KNOWLEDGE BASE

    • How to Start a Podcast
    • How to Start a Live Podcast
    • How to Monetize a Podcast
    • How to Promote Your Podcast
    • Mobile Podcast Recording Guide
    • How to Use Group Recording
    • Podcast Advertising 101
  • Support

    • Support Center
    • What’s New
    • Free Webinars
    • Podcast Events
    • Podbean Academy
    • Podbean Amplified Podcast
    • Badges
    • Resources
  • Podbean

    • About Us
    • Podbean Blog
    • Careers
    • Press and Media
    • Green Initiative
    • Affiliate Program
    • Contact Us
  • Privacy Policy
  • Cookie Policy
  • Terms of Use
  • Consent Preferences
  • Copyright © 2015-2025 Podbean.com