Rohin Shah comments on What more compute does for brain-like models: response to Rohin

Rohin Shah 13 Apr 2022 7:52 UTC
5 points
total compute needed for training = 3.82e12 flp/n/s * 4.78e8 bio_seconds * 1.3e9 neurons = 2.37e30 flops = 2.37e15 petaFLOPs
Should that be 3.82e15 flp/n/s, based on the numbers right above?
- Nathan Helm-Burger 13 Apr 2022 14:45 UTC
  3 points
  Parent
  Yes, thanks for catching that.
  edit: fixed