• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI
    Announcement

    A new paper proposes a scientific theory of atomic features

    The authors say their experiments found that SAEs of different sizes and training data share many features.

    2 Sources, 12h ago, first seen 12h ago

    TLDR

    The authors say their paper mathematically derives testable predictions about atomic features. They report that SAEs of different sizes and training data share many features, and that large SAEs recover both parent and child features—results they say challenge conventional wisdom.

    Combined views

    —

    2 Sources, first seen 12h ago

    Combined views

    —

    2 Sources, first seen 12h ago

    — likes
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    — likes
    — comments
    — saves
    — reposts
    — comments
    — saves
    — reposts
    Featured Source

    2 Sources

    Kenny Peng@kennylpengOur new paper introduces a scientific theory of atomic features. We mathematically derive testable predictions. Our experiments challenge conventional wisdom. SAEs of different size and training data share many features. Large SAEs recover both parent and child features. 🧵12h
    jessica dai@jessicadai_RT @kennylpeng: Our new paper introduces a scientific theory of atomic features. We mathematically derive testable predictions. Our experim…1h

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    2 Sources

    Kenny Peng@kennylpengOur new paper introduces a scientific theory of atomic features. We mathematically derive testable predictions. Our experiments challenge conventional wisdom. SAEs of different size and training data share many features. Large SAEs recover both parent and child features. 🧵12h
    jessica dai@jessicadai_RT @kennylpeng: Our new paper introduces a scientific theory of atomic features. We mathematically derive testable predictions. Our experim…1h