Back

@jensorensen Yes. But this only applies to basic LLM strategies. It doesn’t apply for Self-Taught-Reasoning systems, STaR, which are behind the latest models that have other models first create synthetic higher quality data for training the final models.