3 ms·
Would this not mean we need much much more training data to fully utilize the now "free" capacities?
by neuronic 2y ago
Would this not mean we need much much more training data to fully utilize the now "free" capacities?
- vonneumannstan 2y agoIt's pretty clear that the reasoning models are using mass amounts of synthetic data so it's not a bottleneck.