If the image synthesis is an example, I believe this to be a winning strategy in the long run to get the most competent model. The LLM space just moves a bit sl

If the image synthesis is an example, I believe this to be a winning strategy in the long run to get the most competent model. The LLM space just moves a bit slower since the entry requirements for compute power are that much higher and data preparation is maybe a bit more "dry" than tagging images.