Product Labs🔬: Decoding GitHub Copilot

To those who don’t know, Spark was a software engineer in a past life before deciding to hit Ctrl+Z! Had GitHub Copilot come out a few years earlier, Spark would still be living the developer life, instead of switching over to the other side of the table and becoming a PM (probably). Trouble, however, is a rockstar data scientist and now is also a GitHub Copilot expert (now we know how he earned the title of rockstar data scientist 😉)

What’s in it for you?

Before even understanding what GitHub Copilot is, what on earth is GitHub? Simply put, GitHub is a platform for software developers to store, share, and collaborate on code projects. Today, there are over 100 million developers on GitHub.

During the COVID lockdown, the team at GitHub gained early access to OpenAI's powerful language model, then known as GPT-3. They experimented by feeding the raw model random programming exercises. To their astonishment, the model solved a staggering 93% of them correctly in the initial attempts. Recognizing the potential of this breakthrough, the team at GitHub swiftly developed the groundbreaking developer tool - GitHub Copilot: an AI assistant that not only predicts and completes code but also helps developers understand snippets, and write tests & docs in a jiffy - a perfect example of Disruptive Innovation.

Now you understand, why it could have helped Spark 😋

GitHub has quickly become one of the most widely adopted AI Developer Tools. 37,000+ businesses have adopted it and 1 in 3 Fortune 500 Companies use it.

Developers using GitHub Copilot complete tasks at an astonishing efficiency of 55% faster than others.

GitHub Copilot analyzes the developer's current file, including comments, existing code structure, and all other open tabs, to predict the next logical step. This translates into real-time suggestions that streamline coding workflows.

Taking a closer look…

GitHub Copilot has 2 main extensions:

1. Copilot: As you type code, Copilot suggests completions for lines or even entire functions. This can be a huge time saver, especially for repetitive tasks or boilerplate code. The suggestions are shown as grey text, which is called ghost text. So now you can just press the Tab key. The user can also partially accept suggestions or even shuffle between suggestions. Copilot can also explain any piece of code.

2. Copilot Chat: What the OG Copilot didn’t offer is a way to interact with it. Copilot Chat, can be thought of as a ChatGPT in your editor. It takes things a step further by providing an interactive chat interface within your development environment. You can ask Copilot questions in natural language about your code.

One small thing that bites us often is how GitHub Copilot in a way circumvents Fitts’s Law.

Though Copilot suggestions come up by default, Copilot Chat is not easily discoverable. The user has to access it via the “View” option and then select “GitHub Copilot Chat”. Also, the Copilot icon is a small icon at the bottom of the code window in Visual Studio.

GitHub Copilot has a host of amazing features, let us give a quick glimpse of a few:

• Works well with (almost) all Programming Languages: No matter if you're coding in Python, Javascript, or something else, Copilot can understand and suggest code snippets in your preferred language.

• AI-powered Real-time Suggestions: As you type, Copilot acts like a super smart auto-complete, constantly analyzing your code and suggesting what might come next based on what you've already written.

  

  A simple comment led to the next line of code being generated

• Code Completion: Forget typing out entire functions. Copilot can often fill those in for you, learning from the partially written functions or lines of code you’ve written and filling in the remainder.

• Analysis & Explanation: Just select any part of the code and select “Explain This” to understand the snippet, no more burying yourself in deciphering undocumented code anymore.

• Testing & Debugging: Generate unit tests or Copilot can even suggest ways to test your code and propose potential fixes based on the error.

• Automatic Documentation: With the “Generate Docs” function, generate seamless documentation for your code.

GitHub Copilot follows the same principles as any AI application. AI tools are probabilistic, not deterministic. The more clarity is given to it, the better the results. Christopher Harrison, Senior Development Advocate at GitHub, explained the building blocks to using Copilot more efficiently.

• Context: Additional information to help GitHub Copilot generate custom suggestions

• Intent: The specific goal you have in mind when creating the prompt

• Clarity: How easy something is to understand

• Specificity: The level of detail about the task you wish to complete

The impact of GitHub Copilot extends beyond individual developers. By streamlining the development process, it allows teams to iterate faster, experiment more readily, and ultimately, deliver innovative products to market quicker.

Not just helping enterprises, GitHub Copilot is helping democratize software development. Now anyone, irrespective of their native language and knowledge of software development, can start writing code and building at least simple applications.

Let’s be realistic as well, GitHub Copilot is the perfect AI pair programmer. It is not a replacement for human intelligence, but rather an augmentation of the same. Additionally, Copilot should NOT be seen as a replacement for thorough code review and testing practices.

What’s the intrigue?

Microsoft has been releasing a host of Copilots and Copilot Studios, GitHub Copilot and Azure AI Studio have been around for quite some time, while the latest one in the family is Copilot Studio.

Maybe it’s just, us maybe it’s everyone who gets confused on how are each of these different and what to use when.

• GitHub Copilot is the AI pair programmer, that streamlines coding by suggesting syntax and code blocks within IDEs. You can use it for rapid prototyping, boilerplate code, and repetitive tasks.

• Azure AI Studio, on the other hand, is a centralized hub for AI engineers to explore, deploy, and manage Azure AI services, focusing on the entire AI project lifecycle. It’s best to use when you need a customized AI solution that goes beyond what’s available out of the box.

• Microsoft Copilot Studio differentiates itself as a low-code platform enabling the creation of custom AI conversational agents, enhancing user interactions across Microsoft’s suite. For example, you can think about building a specialized AI copilot for a specific domain, like retail.

While GitHub Copilot directly aids in code generation, Azure AI Studio and Microsoft Copilot Studio offer broader AI solution development and customization capabilities, respectively. Together, they represent a comprehensive suite of tools that cater to various stages of AI integration and application development.

With the advent of generative AI and the host of developer support tools, including the no-code tools more leaps have been made in computing in the last 2 years than in the last century. Spark and Trouble eagerly waiting to see the next developments in this space and if any of them can inspire Spark back into at least a hobby coder.

Whatcha Got There?!🫣

Buckle up, tech fam! Every week, our dynamic duo “Spark” ✨ & “Trouble”😉 share some seriously cool learning resources we stumbled upon.

You Asked 🙋‍♀️, We Answered ✔️

Question: Recently, there have been a host of big tech events - Microsoft Build, Google I/O and Nvidia’s Computex. What do you think was the most exciting announcement?

Answer: That’s a fun question, there were many super cool announcements. We had exciting favourites in each.

• From Microsoft Build, we learnt what to add to our wish lists - The Copilot + PCs. Copilot+ PCs are supercharged Windows machines with built-in AI smarts. They boast impressive battery life and super-fast processors for tackling demanding tasks. These PCs come with unique features like "Recall" to find past activities and "Cocreator" to generate images on the fly, all powered by on-device AI. They’re set to launch in September this year.

• Google I/O hands down announced Project Astra. It is a universal AI agent that blends the capabilities of Google Assistant and Google Gemini. It’s designed to be helpful in everyday life. Astra can view the world through text, audio, or video inputs, identify objects, explain code, find items, and even suggest names for a dog. Its conversational interface enhances user interactions, making it a pivotal tool in Google’s AI arsenal.

• Nvidia unveiled the Rubin GPU architecture at Computex 2024, but it's not slated for release until 2026. Rubin will succeed the Blackwell architecture which is expected in 2025. It is expected to deliver a significant performance jump compared to Blackwell, likely due to a combination of factors like a new 4x reticle design, TSMC CoWoS-L packaging, and next-generation HBM4 memory. Rubin boasts double the networking speed of its predecessors, reaching a staggering 1600GB/second. This will enable faster communication between GPUs across data centres.