3 ms·
Even assuming that past rates of inference cost scaling hold up, we would only expect a 2 OoM decrease after about a year or so. And 1% of 3.5b is still a very
by achierius 2y ago
Even assuming that past rates of inference cost scaling hold up, we would only expect a 2 OoM decrease after about a year or so.
And 1% of 3.5b is still a very large number.
- popcorncowboy 2y agoAnd to your point "past performance is not indicative of future results". The extrapolate to infinity approach is the mindfever of this field.