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

    Teortaxes Shares Charts on Astra Post-Training

    Teortaxes posts charts and text on flat AI scaling curves.

    MB
    AM
    NI
    10 Sources, 27d ago, first seen 27d ago

    TLDR

    Teortaxes posted that two charts display the same phenomena and that Astra's post-training is still far from done. The account stated the model does not yet reliably turn added compute into better results in a monotonic way. A follow-up reply from the same user attached an image titled "Two readings of the flat curve, and the chart can't distinguish them" that contrasts one view attributed to Teortaxes with another view.

    Combined views

    66.6K

    10 Sources, first seen 27d ago

    Combined views

    66.6K

    10 Sources, first seen 27d ago

    1.3K likes
    1.3K likes
    17 comments
    113 saves
    19 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    17 comments
    113 saves
    19 reposts

    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

    10 Sources

    @teortaxesTexThese charts are the same phenomena Do you notice the craziest thing, anon? Astra's post-training is still far from done It doesn't know how to reliably turn more compute into more results. Doesn't know yet.
    @andersonbcdefg@teortaxesTex meh i think this just shows arc agi 3 more trainable via rl envs than its creators would want u to believe lol
    @nickcammaratawoke up, bleary eyes saw this and im like wow it bends early they must not have figured out reasoning yet then i realized it bent bc it finished it
    @aidan_mclau@teortaxesTex i feel like deepseek could def get this
    @sarthakghOpen models these days
    @Miles_BrundageRT @aaronscher: fascinating new compute scaling paradigm we find ourselves in (satire)

    10 Sources

    @teortaxesTexThese charts are the same phenomena Do you notice the craziest thing, anon? Astra's post-training is still far from done It doesn't know how to reliably turn more compute into more results. Doesn't know yet.
    @andersonbcdefg@teortaxesTex meh i think this just shows arc agi 3 more trainable via rl envs than its creators would want u to believe lol
    @nickcammaratawoke up, bleary eyes saw this and im like wow it bends early they must not have figured out reasoning yet then i realized it bent bc it finished it
    @aidan_mclau@teortaxesTex i feel like deepseek could def get this
    @sarthakghOpen models these days
    @Miles_BrundageRT @aaronscher: fascinating new compute scaling paradigm we find ourselves in (satire)