This year has seen an unprecedented wave of Chat GPT (Chat Generative Pre-trained Transformer), an artificial intelligence chatbot program. The emergence of Chat GPT has brought about a monumental shift in the tech industry, as everyone eagerly anticipates witnessing the capabilities of Chat GPT. Never before had we imagined that AI (Artificial Intelligence) could exhibit behavior so close to that of a human. Now, Chat GPT has demonstrated this possibility to the entire world, igniting fervent discussions on how to effectively leverage AI and coexist with it. Riding on this wave of AI, today we are going to discuss Azure's Azure OpenAI service.

Azure OpenAI: A Collaboration Between Microsoft and OpenAI

Azure OpenAI, jointly developed by Microsoft Azure and the acclaimed OpenAI team known for creating Chat GPT, is an AI service platform based on the GPT architecture created by OpenAI. It offers users a diverse range of functionalities, along with robust performance and high scalability for an enhanced AI experience.

Within Azure OpenAI, several key concepts are frequently mentioned. Let's explore each of them:

• Prompts & Completions: The API provides access to the text input and output interface of models. When users enter commands (Prompts), the responses they receive are Completions.

• Tokens: Tokens are primarily divided into Text Tokens and Image Tokens:

1. Text Tokens: Azure OpenAI breaks down text into tokens to process the received text. The form of tokens may be a single word or a block of characters within a word. The total number of tokens depends on the user's input and corresponding output, as well as the requested parameter length, which affects the model's response.

2. Image Tokens: The size of the image and the detailed data of the image are the two main indicators that determine the total number of image tokens. Low-resolution data allows for fewer tokens, resulting in faster API responses. Basic mapping requires 85 tokens. High-resolution data, on the other hand, involves detailed image recognition, where the API crops the image into smaller squares, and each square uses multiple markers to generate text. The total number of tokens is calculated through a series of adjustment steps and tends to be larger.

• Deployments & Models: After creating Azure OpenAI resources in Azure, users need to deploy models. Different models have different functionalities and billing methods. Models like the well-known GPT-3.5 and GPT-4 are used for understanding and generating natural language, while the DALL-E model is used to generate images through natural language. Once deployed, users can start using APIs to call and generate text.

Incredibly Powerful Features of Azure OpenAI

Azure OpenAI currently offers three major AI functionalities, as explained below:

• Natural Language Generation: The GPT model behind natural language generation is either GPT-3.5 or GPT-4. When users provide natural language Prompts, the model can generate corresponding responses based on the Prompts. The GPT model is suitable for many different task types, including summarizing text, classifying text, generating names, translating, and suggesting content, as illustrated in the example provided by the official documentation.

GPT 模型可進行的自然語言工作型態列舉
（圖片取自 Azure 官方教學文檔：https://learn.microsoft.com/zh-tw/training/modules/explore-azure-openai/5-understand-openai-natural-language）

• Code Generation: The GPT model is proficient in many programming languages such as Python, JavaScript, C#, and PHP. It can provide responses to code writing based on natural language Prompts and even conduct unit tests for the code, as shown in the example provided by the official documentation.

• Image Generation: The model used for image generation is DALL-E, which mainly functions in three categories: image creation, image editing, and image transformation. Users can provide Prompts to request AI to generate images or directly provide an image for AI to make changes. Examples are provided in the official documentation.

	Image Generation Result Given "An elephant standing on its hind legs with a hamburger on its head"
	Image Editing Result Given "A blue gorilla reading a book".
	Continuing from the first example, the desired level of change, such as color, appearance, etc., was given, resulting in the image transformation result.

（表格圖片皆取自 Azure 官方教學文檔：https://learn.microsoft.com/zh-tw/training/modules/explore-azure-openai/7-understand-openai-image-generation）

Leading Advantages of Azure OpenAI

So, what are the advantages of using Azure OpenAI? Let's take a look:

• Powerful Performance: As mentioned earlier, Azure OpenAI is built on the GPT model from OpenAI, which provides robust technical support. Additionally, Microsoft Azure offers a cloud environment that allows users to easily use AI services on Azure.
• Superior Scalability: Azure OpenAI can provide corresponding computing power according to the user's needs, with flexible scaling up and down to accommodate projects of various sizes. This not only enhances performance but also reduces costs.
• High Security and Compliance: Data transmission can be encrypted, and data transmission security and stability can be enhanced through private networks or private key encryption. Access controls can be strengthened through identity management with Azure AD and MFA, and strict content moderation can be enforced through Azure Content Moderation filters to filter out negative content, such as offensive content, and perform manual reviews when necessary.

Azure 官方文檔 - 〈什麼是 Azure OpenAI 服務？〉https://learn.microsoft.com/zh-tw/azure/ai-services/openai/overview
Azure 官方教學文檔 - 〈了解 OpenAI 的自然語言功能〉https://learn.microsoft.com/zh-tw/training/modules/explore-azure-openai/5-understand-openai-natural-language
Azure 官方教學文檔 - 〈了解 OpenAI 程式碼產生功能〉https://learn.microsoft.com/zh-tw/training/modules/explore-azure-openai/6-understand-openai-code-generation
Azure 官方教學文檔 - 〈了解 OpenAI 的影像產生功能〉https://learn.microsoft.com/zh-tw/training/modules/explore-azure-openai/7-understand-openai-image-generation