Github Copilot: Good or Bad?

TL;DR

GitHub Co-pilot is an in-editor extension that suggests code while you write, and despite privacy and job concerns, it proves to be incredibly useful and innovative.

Transcript

what is going on everybody and welcome to a video on github co-pilot so there's been a lot of hype and uh dangers expressed with co-pilot and it's been hard for me because i haven't had access until a few weeks ago so i finally have had some time to play around with co-pilot and i thought i would share my thoughts with you all so if you're been liv... Read More

Key Insights

• 👨‍💻 GitHub Co-pilot is an innovative tool that suggests code while programmers write, enhancing productivity.
• 👨‍💻 It is built using OpenAI's Codex model, which focuses specifically on coding, making it a valuable tool for programmers.
• 👨‍💻 Co-pilot's understanding of math and ability to generate accurate suggestions for regular expressions showcases its deep understanding of code patterns.
• 👨‍💻 While there are concerns about privacy and job displacement, Co-pilot is seen as an additional layer of productivity and a tool to learn new coding techniques.

Questions & Answers

Q: How does GitHub Co-pilot work?

GitHub Co-pilot is an in-editor extension that suggests code while you write. It uses OpenAI's Codex model to generate these suggestions, which are based on the code you're working on.

Q: What is the difference between Codex and GPT-3?

Codex is a smaller model with 12 billion parameters compared to GPT-3's 175 billion parameters. Codex is focused on coding, while GPT-3 is a general-purpose model that includes various concepts.

Q: Can GitHub Co-pilot handle large code files?

Co-pilot has a context size limit of 14 kilobytes. So, while it can handle considerable code files, it may ignore some earlier or further code when generating suggestions.

Q: Does GitHub Co-pilot understand regular expressions?

Yes, Co-pilot can assist in writing regular expressions. It can even anticipate the output of regular expressions based on specific pattern matching and provide accurate suggestions.

Q: Will GitHub Co-pilot replace programmers and take away jobs?

Co-pilot is more like an extra layer of abstraction, similar to using a higher-level programming language. It enhances productivity and makes programmers more valuable, ultimately creating more job opportunities.

Q: Can GitHub Co-pilot teach programmers new coding techniques?

Yes, Co-pilot can introduce new coding techniques and libraries to programmers. It can guide them through unfamiliar code and suggest alternative methods they may not have thought of before.

Q: Does GitHub Co-pilot make mistakes?

Co-pilot, like any AI model, can make errors in understanding the desired outcome. However, there is a skill in prompting Co-pilot effectively, similar to using search engines to find accurate information.

Q: What are the overall impressions of GitHub Co-pilot?

Despite initial skepticism, the author found GitHub Co-pilot to be incredibly helpful and productive. It integrates well into the coding workflow and produces accurate suggestions quickly.

Summary & Key Takeaways

• GitHub Co-pilot is an extension that suggests code and can even generate entire blocks of code while you write.

• It was created using OpenAI's Codex model, which is smaller than GPT-3 but more focused on coding.

• Co-pilot has a context length limit of 14 kilobytes, can understand math, and shows a deep understanding of code patterns.