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

    Will Depue Highlights Strong Vision Evals

    AI engineer notes strong vision benchmark results for an unspecified model.

    DE
    WD
    BP
    6 Sources, 25d ago, first seen 25d ago

    TLDR

    Will Depue posted that a team achieved something unusual with perception in one model. He wrote that all the vision evals are off the charts yet added he has no idea what produced the gains. Boris Power replied that vision works remarkably well and makes Astra more useful for real-world tasks. Both posts quote the same source tweet. The statements remain unconfirmed beyond these attributed social posts and name no specific model or benchmark numbers.

    Combined views

    69.5K

    6 Sources, first seen 25d ago

    Combined views

    69.5K

    6 Sources, first seen 25d ago

    837 likes
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    837 likes
    31 comments
    148 saves
    24 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    31 comments
    148 saves
    24 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    6 Sources

    @willdepuethey did something magical for the perception of this model. no idea what but all the vision evals are off the charts
    @BorisMPowerVision works remarkably well. This makes Astra a lot more useful for tasks interfacing with the real world
    @teortaxesTexGoogle has not a single moat left, huh long context is dominated by OAI/Anthropic and many Chinese models handle it fine and for cheaper. Vision – Anthropic is catching up, Qwen is on par, Astra ahead. Core LLM… lol. Updating 3.x Flash won't cut it guys
    @deedydas[Vision] Variety of real-world vision tasks from the Roboflow team 4/22
    @braddwyerAstra is now the default model for auto-labeling in @roboflow.
    @peteskomorochRT @braddwyer: Astra is now the default model for auto-labeling in @roboflow.

    6 Sources

    @willdepuethey did something magical for the perception of this model. no idea what but all the vision evals are off the charts
    @BorisMPowerVision works remarkably well. This makes Astra a lot more useful for tasks interfacing with the real world
    @teortaxesTexGoogle has not a single moat left, huh long context is dominated by OAI/Anthropic and many Chinese models handle it fine and for cheaper. Vision – Anthropic is catching up, Qwen is on par, Astra ahead. Core LLM… lol. Updating 3.x Flash won't cut it guys
    @deedydas[Vision] Variety of real-world vision tasks from the Roboflow team 4/22
    @braddwyerAstra is now the default model for auto-labeling in @roboflow.
    @peteskomorochRT @braddwyer: Astra is now the default model for auto-labeling in @roboflow.