But it's still only 50 something percent on HumanEval right? So to really prove this out you still need to try to make a much larger model.
I wish someone would do something like that but not use OpenAI's models to create the training data because then supposedly it can't be used for commercial purposes.
Not that it's easy to create a lot of high quality training data.
I wish someone would do something like that but not use OpenAI's models to create the training data because then supposedly it can't be used for commercial purposes.
Not that it's easy to create a lot of high quality training data.