3.5.4 Fix This Program Github

[Rosicler Kleckner]

NoteIf you like to learn about the latest GitHub Copilot experience in Visual Studio Code, go to the GitHub Copilot in VS Code topic, where you'll find details on the Copilot editor integration and Copilot Chat features such as inline Chat, the dedicated Chat view, and Quick Chat.

There is a lot of buzz, excitement, and some concerns around Artificial Intelligence today. Advancements are happening almost daily, it's hard to keep up. But once you give it a try, you quickly realize what more than a million Copilot users see daily, that this technology does not disappoint, especially with Large Language Models (LLMs) like OpenAI's GPT-3.5/4.

It's been more than a year since the Copilot extension first launched. Oege De Moor and his team at GitHub collaborated with OpenAI to bring the first at scale example of using LLMs for development, specifically OpenAI's codex model.

Copilot offers real-time hints for the code you are writing by providing suggestions as "ghost text" based on the context of the surrounding code. If you write your intentions as comments or use descriptive function names, Copilot can use these hints to automatically generate code, word by word, line by line, or even block by block.

We knew it was just a matter of time before AI permeated the development process. We had been working with the GitHub Next team on Copilot and the Copilot Labs extension, but the release of ChatGPT in November 2022 significantly accelerated the timetable.

The VS Code team had an internal hackathon to explore how to integrate AI into VS Code more deeply and came up with many cool ideas such as improved renaming and refactorings, code transformations based on examples, and ways to create file glob patterns or regex expressions using natural language.

As interesting as these ideas were, we kept circling around the Chat experience. We had a healthy dose of skepticism. Did we really need yet another Chat experience in our lives? VS Code is an editor, our job is to get you into the "zone" and help you stay there by removing distractions while you crank out the next unicorn.

In the example below, we first selected the users array and then pressed Cmd+I on macOS (or Ctrl+I on Windows/Linux) to bring up the Chat input. We then asked Copilot to split the username field into both firstName and lastName, and last we clicked on Inline Diff to see the changes Copilot made.

No need to express your intentions to Copilot using a comment or an overly descriptive function name, just press Cmd+I and describe what you want to do. Stay in the zone while the Copilot does the work.

As the excitement around ChatGPT grew, so did the number of internal and external extensions with custom Chat views. We worried that these would not scale, it's hard to support the basics like key bindings and themes in a web view, and for those using the Monaco editor for code blocks, it is very hard to scale to hundreds of instances, not to mention there is no extension support in those editor instances. As a result, we partnered closely with the GitHub Copilot team to build a Chat view into the core of VS Code.

The most obvious value of an integrated experience over a browser-based chat is that we can provide context to the model, helping to ground the conversation and provide better answers. For example, you can't easily ask the browser-based ChatGPT to optimize code that spans multiple files. VS Code already knows a lot about the workspace as this is how you can do refactorings across files, Find All References, Go to Definition, and more. Key information can be responsibly embedded in the prompt so that the model can give more relevant answers, ultimately letting you ask Copilot to optimize or refactor code that has dependencies across multiple files.

Answers to many programming questions often include multiple steps. Think about how many blog posts walk you through a process - a Chat view is well suited for this type of interaction. A great example is debugging. If we're being honest (and we are), it isn't easy to configure launch.json and tasks.json for debugging. Instead of us writing more documentation or adding dialogs and wizardry to the editor, it turns out you can simply ask "/vscode add a launch.json and a tasks.json" which then provides instructions on how to create the files, content to add, and so on, until you are hitting breakpoints and inspecting variables like a boss.

But perhaps the most important reason for integrating chat is because having a two-way conversation helps you decide what is right and what is wrong. Large language models are not perfect, and they don't "think". They simply figure out the next best word to respond with (granted, they are pretty good at this).

As the Pilot, you are always in charge, and you decide which of Copilot's suggestions to take and what code to bring into your workspace. The ability to ask clarifying questions or provide additional specifics helps you make those key decisions.

Naming is hard (we should know, our product is called "Code"). However, "GitHub Copilot", is a great name. Rather than "pair programmer" or "auto pilot" or simply "chat", "Copilot" conveys multiple important concepts in a single word.

Monumental advancements in technology like this really are few and far between. We too believe AI is the next big shift that will transform how we think about, build, and use developer tools. It will ultimately enhance every aspect of the development process in ways we can only dream about today. We don't claim to be the first to say this, but soon we will all wonder how we ever built, debugged, deployed, and maintained applications and systems without AI powered tools.

AI is not perfect (neither are we!) and it will improve over time. Microsoft and GitHub Copilot follow Responsible AI principles and employ controls to ensure that your experience with the service is appropriate, pleasant, and useful. We understand there is hesitation and concern surrounding the rapid expansion of AI's capabilities, and fully respect those who don't want or can't use Copilot.

The combination of a Chat assistant, in-editor Chat, and inline completions let you leverage the power of Copilot no matter what coding "zone" you are in. Just getting started or have questions? Use the Chat view. Furiously writing code and want to do some refactoring, updating, or scaffolding? Smash Ctrl+I and ask Copilot to do it for you. Working your way through a problem step by step? Use inline completions to Tab-Tab-Tab your way to success.

Integrating Chat into VS Code, powered by GitHub Copilot, is just one step on a journey. We will continue to explore additional "inner loop" scenarios such as using natural language to describe search patterns or code generation, automatic generation of commit messages and pull request descriptions, smarter code renaming, refactoring, transformations, and more.

The GPT-3.5-Turbo and GPT-4 models are language models that are optimized for conversational interfaces. The models behave differently than the older GPT-3 models. Previous models were text-in and text-out, which means they accepted a prompt string and returned a completion to append to the prompt. However, the GPT-3.5-Turbo and GPT-4 models are conversation-in and message-out. The models expect input formatted in a specific chat-like transcript format. They return a completion that represents a model-written message in the chat. This format was designed specifically for multi-turn conversations, but it can also work well for nonchat scenarios.

This article walks you through getting started with the GPT-3.5-Turbo and GPT-4 models. To get the best results, use the techniques described here. Don't try to interact with the models the same way you did with the older model series because the models are often verbose and provide less useful responses.

The following code snippet shows the most basic way to use the GPT-3.5-Turbo and GPT-4 models with the Chat Completion API. If this is your first time using these models programmatically, we recommend that you start with the GPT-3.5-Turbo and GPT-4 quickstart.

In the Azure OpenAI documentation, we refer to GPT-3.5-Turbo and GPT-35-Turbo interchangeably. The official name of the model on OpenAI is gpt-3.5-turbo. For Azure OpenAI, because of Azure-specific character constraints, the underlying model name is gpt-35-turbo.

Unlike previous GPT-3 and GPT-3.5 models, the gpt-35-turbo model and the gpt-4 and gpt-4-32k models will continue to be updated. When you create a deployment of these models, you also need to specify a model version.

OpenAI trained the GPT-35-Turbo and GPT-4 models to accept input formatted as a conversation. The messages parameter takes an array of message objects with a conversation organized by role. When you use the Python API, a list of dictionaries is used.

The system role, also known as the system message, is included at the beginning of the array. This message provides the initial instructions to the model. You can provide various information in the system role, such as:

To trigger a response from the model, end with a user message to indicate that it's the assistant's turn to respond. You can also include a series of example messages between the user and the assistant as a way to do few-shot learning.

The following section shows examples of different styles of prompts that you can use with the GPT-35-Turbo and GPT-4 models. These examples are only a starting point. You can experiment with different prompts to customize the behavior for your own use cases.

You can also include relevant data or information in the system message to give the model extra context for the conversation. If you need to include only a small amount of information, you can hard code it in the system message. If you have a large amount of data that the model should be aware of, you can use embeddings or a product like Azure AI Search to retrieve the most relevant information at query time.

3a8082e126