From what I can tell from the preview, this is right. I don’t have access to gpt-3 to compare, but copilot seems to have the same functionality that gpt-3 advertises when editing a plain text file.
> Since Codex is evaluated on natural language prompts, we hypothesized that it would be beneficial to fine-tune from the GPT-3 (Brown et al., 2020) model family, which already contains strong natural language representations. Surprisingly, we did not observe improvements when starting from a pre-trained language model, possibly because the finetuning dataset is so large. Nevertheless, models fine-tuned from GPT converge more quickly, so we apply this strategy
for all subsequent experiments.
It's not completely clear exactly what relationship the Codex models and Copilot have to one another, but given that the Copilot model is internally named "Cushman" (going by the API URL), which is the same name as the faster of OpenAI's two Codex models, they're probably trained the same way.
