Comments by "Michael Nurse" (@michaelnurse9089) on "New Mind" channel.

  1.  @richharr  Some?
    1
  2. It is a mostly irrelevant problem as the lack of good data is a 10x bigger problem. In my somewhat limited experience most neural nets train pretty quickly. Even groundbreaking algorithms like Alpha Zero trained in only 4 hours. Sure there are some big training requirements like GPT-3 but for those, creating and cleaning the data was a MUCH more expensive and time consuming exercise. Nvidia stills sells more cards for graphics than for compute revenue wise .
    1