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

    DeepMind Scientist Flags Early $1bn Model Discussions

    Reply by DeepMind scientist David Pfau on old AI scaling talks.

    DP
    T(
    DF
    12 Sources, 26d ago, first seen 26d ago

    TLDR

    David Pfau, Staff Research Scientist at Google DeepMind and visiting professor at Imperial College London, posted a reply noting the dated nature of a discussion. Pfau wrote that the exchange felt like a time capsule because participants were already talking about training a $1bn model three years earlier. The comment stands as a first-party observation from Pfau on the age of the referenced conversation. No further details on the original post or additional replies appear in the supplied evidence.

    Combined views

    38.6K

    12 Sources, first seen 26d ago

    Combined views

    38.6K

    12 Sources, first seen 26d ago

    406 likes
    406 likes
    15 comments
    47 saves
    14 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    15 comments
    47 saves
    14 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

    12 Sources

    @pfauThis was for the pretty simple reason that $10bn is about the most any industry will spend on any single artifact. It's nuclear power plant/Saturn V/CERN/offshore oil rig territory.
    @sytelus@pfau This bound assumes traditional view that investments follow expected ROI which is not true in AI because ROI is assumed to be infinite.
    @DanielleFong100k GPUs is the new Heavyweight class. I wonder what the Super Heavyweight will look like
    @yishan@DanielleFong DGX Station?
    @adeen14ai@somewheresy 100k in one building and it's the biggest release of the year. the sites being built right now are ten times that. that's the part that should make people uncomfortable, not this one
    @somewheresythis was written by an AI but it is correct
    @teortaxesTexIt wasn't even 2% of world's compute. Maybe 1%. An interesting thought experiment: imagine if instead of American tech startups building it, we'd have had Astra as an explicit international scientific project, a la ITER or the ISS. What's the reasonable budget?
    @zehavocRT @adeen14ai: @somewheresy 100k in one building and it's the biggest release of the year. the sites being built right now are ten times th…

    12 Sources

    @pfauThis was for the pretty simple reason that $10bn is about the most any industry will spend on any single artifact. It's nuclear power plant/Saturn V/CERN/offshore oil rig territory.
    @sytelus@pfau This bound assumes traditional view that investments follow expected ROI which is not true in AI because ROI is assumed to be infinite.
    @DanielleFong100k GPUs is the new Heavyweight class. I wonder what the Super Heavyweight will look like
    @yishan@DanielleFong DGX Station?
    @adeen14ai@somewheresy 100k in one building and it's the biggest release of the year. the sites being built right now are ten times that. that's the part that should make people uncomfortable, not this one
    @somewheresythis was written by an AI but it is correct
    @teortaxesTexIt wasn't even 2% of world's compute. Maybe 1%. An interesting thought experiment: imagine if instead of American tech startups building it, we'd have had Astra as an explicit international scientific project, a la ITER or the ISS. What's the reasonable budget?
    @zehavocRT @adeen14ai: @somewheresy 100k in one building and it's the biggest release of the year. the sites being built right now are ten times th…