Reaction
Distillation's possible role in China's AI model benchmark performance
Another poster proposes two full post-training runs, one using raw reasoning from closed models and the other open models.
TLDR
A poster who recently joined OpenAI says distillation has long been part of the explanation for why China is close behind the U.S. on model benchmarks, but now thinks it probably plays a bigger role than he'd realized. Another poster wants a paper comparing two runs of a full post-training pipeline: one using raw reasoning from closed models and the other using open models. That poster estimates the work would need major lab coordination and low-to-mid eight figures for compute and environments.
Combined views
—
1 Source, first seen ago
