4 ms·
>Ultimately, R1-Zero demonstrates the prototype of a potential scaling regime with zero human bottlenecks – even in the training data acquisition itself. I wou
by artninja1988 2y ago
>Ultimately, R1-Zero demonstrates the prototype of a potential scaling regime with zero human bottlenecks – even in the training data acquisition itself.
I would like this to be true, but doesn't the way they're doing RL also require tons of human data?
- Davidzheng 2y agoI think yes. But hopefully in math with compute advances we can lower the human data input by increasing the gap that is bridged by raw model capabilities vs search augmentation (either with tree search or full rollouts)