Copilot is GitHub’s newly released “AI pair programming tool” that caused quite a stir upon its release. I was fortunate to gain access to the preview version for testing, and this blog post aims to share my initial thoughts and experiences with this tool. In short, I was blown away by its powerful capabilities. After playing with it for a few hours, I found myself genuinely amazed several times, thinking that your development tools could never achieve this level!
However, this tool currently has many limitations in practice, which I will discuss in detail in this article. To sum up:
-
Copilot can appear unexpectedly, interrupting my train of thought
-
The autocomplete feature doesn’t feel like genuine “pair programming”; the interaction mode seems to be off
-
Checking Copilot’s work increases cognitive load
My feeling is that this tool currently won’t revolutionize programming. Nevertheless, despite the aforementioned issues, I still believe it will have a significant, game-changing impact in the future.
If you’re interested in learning more, please continue reading…
Copilot is the latest development from OpenAI, a San Francisco-based artificial intelligence company that recently received substantial support from Microsoft (1 billion USD). OpenAI’s recent GPT-3 (Generative Pre-trained Transformer 3) has made headlines, being their third-generation language model, based on a large-scale neural network with 175 billion parameters, trained on a massive text corpus.
To quote Arram Sabeti:
I accessed the OpenAI GPT-3 API and was truly blown away. It is more fluent and natural than any AI language system I’ve tried.
GPT-3 can generate very fluent and natural prose based on simple prompts. Additionally, many have developed interesting tools through their API, including question-based search engines, chatbots that allow you to converse with historical figures, and even generating guitar tabs. Lastly, Sharif created an impressive demo that can create HTML layouts based on text prompts—this already hints at the embryonic form of Copilot.
Copilot is based on Codex, a derivative model of GPT-3, which has been trained on a vast amount of open-source code from GitHub. It integrates directly with VSCode and can generate suggestions based on a combination of the current context (i.e., your code) and the