Azure OpenAI

Azure OpenAI

Early this week Satya Nadella, Microsoft CEO, and Eric Boyd, Corporate AI Vice President, announced Azure OpenAI Service generally available, which will soon include ChatGPT – the fine-tuned version of GPT-3.5 built upon Azure AI infrastructure gone viral in the last few weeks. But let’s take a step back. What is Azure OpenAI?

The Azure OpenAI Service is the result of a 4-years partnership between Microsoft and OpenAI with the mission of building an artificial general intelligence (AGI) system, democratizing AI and generating a broadly distributed economic benefits.

From a technical point of view, Azure OpenAI provides access to OpenAI’s powerful language (like GPT-3.5 and Codex) and computer vision models (like DALL•E 2), through REST APIs, Python Sdk or a web-based interface (Azure OpenAI Studio), by ensuring Azure security and responsible AI capabilities.

Key concepts of how OpenAI generative AI models work are prompt programming and in-context learning.  A prompt is a text command input that is used by the model to generate a text completion. For example, if we ask Codex to generate a Python function to order a list of items, the text used as instruction is the prompt, while the output code – predicted as the most probable next piece of text - is the completion. In addition to the prompt, the user can provide examples of the expected outcome as input: this is what is defined as in-context learning.

How are we using OpenAI today?

Azure OpenAI powers several Microsoft apps and experiences today. An example is GitHub Copilot, an AI pair programmer based on OpenAI Codex model, capable of generating code from a comment, generating documentation from a piece of code or translating between a coding language to another.

Another great example is the Microsoft Designer App, currently in preview, based on DALL•E model and capable of generating images and artworks from textual input.

How can you get started?

If you want to get started generating text, artworks or code through Azure OpenAI services, you need to follow the steps below:

• Create an Azure Subscription. You can activate a free trial and if you are a student you can get additional benefits.
• Apply for access to the service, by completing the form at https://aka.ms/oai/access. To limit risk of misuse, Azure OpenAI currently requires registration and is only available to managed customers and partners committing for using the service responsibly (both for experimentation or production).
• Create an Azure OpenAI resource, by filling in all the required information, as per screenshot below, adjusting subscription name and region according to your needs.
  
• Create a new model deployment, by clicking on Model Deployments -> Create and then selecting the base model you want to test. For text generation, you can start with text-davinci-002.
  

• Grab your key and  base endpoint from the Keys and Endpoint tab 
  

• Consume your model through the REST API, sending a POST request to https://your_endpoint/openai/deployments/your_deployment_name/completions?api-version=2022-12-01 and adding your prompt in the request body.

• Learn and achieve more, by checking those useful quickstart tutorials!