4 ms·

thanks for the paper link! Im surprised there is such a minimal improvement in structured outputs when using any of these tools over the bare LLM!