Cheap hardware and reinforcement learning could speed AI progress, one post argues
The post argues that once the data problem is somewhat solved, post-training and recursive improvement become the bottleneck again.
TLDR
A user argues that once AI’s data problem is somewhat solved, the bottleneck returns to post-training—improving models after initial training—and recursive improvement. They suggest cheap, standardized hardware paired with modern reinforcement learning methods could accelerate progress much faster than people expect.
Combined views
46
1 Source, first seen 23d ago