4 ms·

Man if only computing got cheaper year to year, oh well!

• sanderjd 4y ago
  It doesn't, really, for these big ML models. Most of the cost savings, and thus progress, in recent years has come from innovations in the algorithms. It remains very expensive to run ML systems at huge scale. Everyone is betting (for example, Microsoft through its billions subsidizing OpenAI) that either hardware or software breakthroughs will make this all cost effective over time. That bet may well be right. All I'm saying is that it isn't obvious to me that it will be, and there's this entire mini-boom of companies is dependent on how that gamble turns out.
  • __pache__ 4y ago
    > It doesn't, really, for these big ML models. Hard to take this seriously, sorry man.
    • sanderjd 4y ago
      Why? Because you just blindly assume computing power is always cheap and have trouble with that assumption being challenged? The whole reason chatgpt burst on the scene years after the techniques it is based on were developed is that the incumbents were unwilling to take the hit to their margins by releasing a fundamentally more expensive product category. OpenAI and Microsoft decided to take the gamble on that. Maybe it will pan out, maybe it won't. That's how gambles work.