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

    A proposal to program AI model weights with natural language

    A researcher says precise human direction kept AI-generated suggestions from derailing work on the proposed system.

    Minh Nhat Nguyen 🦭MN
    1 Source, 1h ago, first seen 1h ago

    TLDR

    A researcher working on the project says AI models can produce plenty of research ideas, but human discretion remains valuable. They say the work required precise direction to avoid plausible suggestions that did not move it forward. Their linked draft proposes turning natural-language requests into changes to an AI model’s weights; the team is currently testing the architectural premise, not presenting a working editor.

    Combined views

    142

    1 Source, first seen 1h ago

    Combined views

    142

    1 Source, first seen 1h ago

    2 likes
    2 likes
    1 comments
    2 saves
    1 comments
    2 saves

    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.

    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    1 Source

    Minh Nhat Nguyen 🦭@menhguinanyway this is the pretraining thingy i work on (text to weights using sparse MoE), hmu if ur interested. in this kind of direction, i had to intentionally steer the model's ideas and articulate what im trying to do in very specific and precise terms, or else it'll introduce a dozen half-plausible suggestions that don't necessarily get me any further. and a bunch of pipelines or whatever. i definitely feel it was not as simple as I expected/what people would think. additionally, it's very interesting that i havent seen much slop pretraining papers in recent years. probably because the slop factories wouldn't go for pretraining bc it's just high effort enough. https://docs.google.com/document/d/1HY9NNkvTLWYqxAWYe2oIAxmHeVr5citUTkAozeiBYiw/edit?usp=drivesdk1h

    1 Source

    Minh Nhat Nguyen 🦭@menhguinanyway this is the pretraining thingy i work on (text to weights using sparse MoE), hmu if ur interested. in this kind of direction, i had to intentionally steer the model's ideas and articulate what im trying to do in very specific and precise terms, or else it'll introduce a dozen half-plausible suggestions that don't necessarily get me any further. and a bunch of pipelines or whatever. i definitely feel it was not as simple as I expected/what people would think. additionally, it's very interesting that i havent seen much slop pretraining papers in recent years. probably because the slop factories wouldn't go for pretraining bc it's just high effort enough. https://docs.google.com/document/d/1HY9NNkvTLWYqxAWYe2oIAxmHeVr5citUTkAozeiBYiw/edit?usp=drivesdk1h