LLMs act on probabilities, which means their output will resemble the greater volume of code that's public domain. So it is primarily mediocre to bad.