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

    Research talent versus compute in frontier AI breakthroughs

    A user argues that labs with roughly 10,000 B300s could pursue breakthroughs without much loss of experimental signal.

    kacheKA
    ar0cket1AR
    2 Sources, 1h ago, first seen 1h ago

    TLDR

    A user argues that a lab with around 10,000 B300s could be roughly tied with frontier labs, even if its experiments were half the scale or less. They doubt that smaller experiments above 100 billion parameters would lose much signal, and claim a lab could out-research OpenAI or Anthropic in a promising direction if it has more strong researchers focused on it.

    Combined views

    1.7K

    2 Sources, first seen 1h ago

    16 likes2 comments7 saves3 reposts

    Combined views

    1.7K

    2 Sources, first seen 1h ago

    16 likes2 comments7 saves3 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Featured Source

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    2 Sources

    ar0cket1@ar0cket1Along this line of reasoning even as a frontier lab you don’t really need all that much compute to make breakthroughs. Once you have around the compute which neolabs have (like 10k B300s) your basically tied with frontier labs. Maybe the difference is that every experiment you do is half the scale or less than a frontier lab but above like 100B params I doubt scaling laws change experimental results so you have minimal loss in signal with smaller experiments. All that matters is direction is how much talent is going down that direction. It’s easy to out research OpenAI/anthropic if you have more good researchers on one direction than them as long as the direction is good.1h
    kache@yacineMTBRT @ar0cket1: Along this line of reasoning even as a frontier lab you don’t really need all that much compute to make breakthroughs. Once…1h
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI

    2 Sources

    ar0cket1@ar0cket1Along this line of reasoning even as a frontier lab you don’t really need all that much compute to make breakthroughs. Once you have around the compute which neolabs have (like 10k B300s) your basically tied with frontier labs. Maybe the difference is that every experiment you do is half the scale or less than a frontier lab but above like 100B params I doubt scaling laws change experimental results so you have minimal loss in signal with smaller experiments. All that matters is direction is how much talent is going down that direction. It’s easy to out research OpenAI/anthropic if you have more good researchers on one direction than them as long as the direction is good.1h
    kache@yacineMTBRT @ar0cket1: Along this line of reasoning even as a frontier lab you don’t really need all that much compute to make breakthroughs. Once…1h
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet