Why not advanced models like Claude or GPT4o to generate new data?

Short Course Q&A Improving Accuracy of LLM Applications
1

In the course, the tutor used the llama 3 model to generate data. By default, isn’t larger and more powerful models better options to use to generate new data and then use that data to fine tune small models like Llama 3?

2

@AYYOUB_MANSSOURI I have not taken this class, but my presumption he’s doing it this way is:

  1. Llama 3 is free, in contrast to either Claude or GPT4o.

  2. You can actually download the weights.

P.s. it is also worth noting I would strictly call Llama 3 ‘small’ – It depends on what variant you download/are using.

image
image1964×796 79.8 KB