On-policy distillation trains smaller AI models more efficiently

Original: RT @NielsRogge: One of the hottest terms in AI right now is "On-policy distillation".

Source: x.com ↗