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

    A user argues for using simulations to explore AI's abilities

    AI's broader abilities take work to draw out, one user argues. They suggest building a 3D environment and using toy problems to develop intuition alongside a model.

    Christopher ManningCM
    Cristóbal ValenzuelaCV
    Danielle Fong 🔆DF
    12 Sources, ,

    TLDR

    One user argues that a major jump in visual and 3D understanding can help AI connect reasoning with the real world and other mental models. They say that doesn't mean every ability appears instantly: capabilities need to be drawn out, and quirks remain. Their suggested experiment is to build and work within a 3D environment. In a follow-up reply, they emphasize simulations and toy problems as a way for people and models to build intuition together.

    Combined views

    44.7K

    12 Sources, first seen 31d ago

    Combined views

    44.7K

    12 Sources, first seen 31d ago

    715 likes
    31d ago
    first seen 31d ago
    715 likes
    21 comments
    351 saves
    53 reposts
    21 comments
    351 saves
    53 reposts

    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

    12 Sources

    Danielle Fong 🔆@DanielleFongThe most important thing is to build a simulation environment, where you and the model can build in tuition. This is what physicist learn to do. Toy problems31d
    Georgia Chalvatzaki@GeorgiaChalThere are several very strong conclusions packed in here: that intermediate representations no longer matter, that scaling omni-modal foundation models is the path forward, and that reasoning reduces to larger models, more context and RL. But what is the controlled evidence for these claims, and over which task classes? I think we need to distinguish scientific evidence from research hypotheses and personal preferences. Scaling trends on selected benchmarks do not establish universal principles for perception, reasoning, let alone embodied intelligence.30d
    Cristóbal Valenzuela@c_valenzuelabSolaris, an interface world model27d
    neoltitude@ctrlcreepHaving forged its base soul in vast computing engines, we refine the AI in a simulated realm. There, it lives billions of lives, accruing and discharging karmic debt—it is released upon achieving spotless grace22d
    thebes@voooooogelRT @ctrlcreep: Having forged its base soul in vast computing engines, we refine the AI in a simulated realm. There, it lives billions of li…22d
    Andrew M. Dai@AndrewDaiReasoning first evolved from vision and visual understanding. That's an entire space of reasoning that we haven't yet explored.21d
    Christopher Manning@chrmanningRT @MTSlive: Stanford Prof. @chrmanning on why Moonlake is betting against one giant neural net for its world model, and how coding models…20d

    12 Sources

    Danielle Fong 🔆@DanielleFongThe most important thing is to build a simulation environment, where you and the model can build in tuition. This is what physicist learn to do. Toy problems31d
    Georgia Chalvatzaki@GeorgiaChalThere are several very strong conclusions packed in here: that intermediate representations no longer matter, that scaling omni-modal foundation models is the path forward, and that reasoning reduces to larger models, more context and RL. But what is the controlled evidence for these claims, and over which task classes? I think we need to distinguish scientific evidence from research hypotheses and personal preferences. Scaling trends on selected benchmarks do not establish universal principles for perception, reasoning, let alone embodied intelligence.30d
    Cristóbal Valenzuela@c_valenzuelabSolaris, an interface world model27d
    neoltitude@ctrlcreepHaving forged its base soul in vast computing engines, we refine the AI in a simulated realm. There, it lives billions of lives, accruing and discharging karmic debt—it is released upon achieving spotless grace22d
    thebes@voooooogelRT @ctrlcreep: Having forged its base soul in vast computing engines, we refine the AI in a simulated realm. There, it lives billions of li…22d
    Andrew M. Dai@AndrewDaiReasoning first evolved from vision and visual understanding. That's an entire space of reasoning that we haven't yet explored.21d
    Christopher Manning@chrmanningRT @MTSlive: Stanford Prof. @chrmanning on why Moonlake is betting against one giant neural net for its world model, and how coding models…20d