Someone reverse engineered copilot and found that it was using the weaker faster OpenAI coding model. So they are not using code-davinci-002 or text-davinci-003 both of which are better at code generation.
Also several groups including OpenAI are racing to advance the state of the art in text (and code) generation. We should anticipate better models being available within the next year or so. In fact I believe some papers within the month or so are an improved SOTA in their contexts.
Try the Codex Playground stuff on the OpenAI site.
Comments
Someone reverse engineered copilot and found that it was using the weaker faster OpenAI coding model. So they are not using code-davinci-002 or text-davinci-003 both of which are better at code generation.
Also several groups including OpenAI are racing to advance the state of the art in text (and code) generation. We should anticipate better models being available within the next year or so. In fact I believe some papers within the month or so are an improved SOTA in their contexts.
Try the Codex Playground stuff on the OpenAI site.