3 ms·

Yes, training is a tiny amount of usage, this is well-known.

• MrMcCall 2y ago
  Well, if that's the case, then the situation is worse than I imagined. I thought that training the model would be far more expensive computationally than using it to compute an 'answer'. Thanks.
  • MF-DOOM 2y ago
    Training is much more computationally intensive than computing a _single_ answer. But not compared to running the model over time at scale for millions of users.
    • MrMcCall 2y ago
      Thanks. That is precisely how I intuited the situation. 40+ years of programming FTW!