2 ms·
^ This and we need to be continually learning on an energy budget similar to how much a human spends per hour.
by mattxxx 2y ago
^ This and we need to be continually learning on an energy budget similar to how much a human spends per hour.
- rlupi 2y agoThe main reason why we can't do that now is because we require models to be digitally reproducible (IMHO, but also read Geoffrey Hinton's mortal computing). The energy cost come from error correction as much as training algorithms.