That's an ugly position on Meta's part. But Llama models are small, they are not going to be preferred to generate synthetic data, GPT-4 is the darling of synth datasets.