Likewise, if you can apply current LLMs to identify the worst 10% of your training data and either discard or improve that, the next generation gets a lot better. And you can iterate this

Posters: johnsonmxe. johnsonmxe

4 likes recorded by the source corpus.