Should there be some kind of compensation, in your time estimates, for how many orders of magnitude of compute were available at the time?
In other words, was the progression from below-human to average across these metrics due to more development effort, more compute, or some other variable? You post focuses on algorithmic improvements, but how many of those were doable with the preceding generation’s compute limitations?
Should there be some kind of compensation, in your time estimates, for how many orders of magnitude of compute were available at the time?
In other words, was the progression from below-human to average across these metrics due to more development effort, more compute, or some other variable? You post focuses on algorithmic improvements, but how many of those were doable with the preceding generation’s compute limitations?
All told, great post, and I enjoyed reading it.