Press Space to continue
Post-training systems create durable competitive advantages in AI
Press Space to continue
Post-training systems are the final boss! When successful, you produce higher intelligence per dollar and it builds quite a MOAT because it’s very hard to align all these pieces at once.
dustinvtran@dustinvtranJun 24
RL is the ultimate systems challenge: pretraining's trainer, prod's inference & environment stack, and eval's rewards in one loop trained end-to-end. Whoever runs that loop best at the bleeding edge (nvfp4, 1M context, agentic grading, ...) owns the frontier.