The power of generative AI
AI is not new, in business or everyday life. For example, with Spotify and Netflix, it can predict your preferences based on the music or film that you’ve chosen before. Similarly, the Azure Cognitive Services Suite has had AI functionality for years such as speech to text; text to speech and translation from one language to another as well as face recognition and object detection in videos for example.
The element that’s new with generative AI is content creation. This came to the fore with ChatGPT earlier this year – the fastest growing service to get to 100 million users in just two months.
Generative AI like ChatGPT is powered by large language models such as GPT 3.5 Turbo or GPT 4, and Microsoft services are helping business to harness their immense potential – safely. The beauty of these large language models is that not only do they understand plain English and interpret it into searchable keywords, but they can manipulate and summarise that text. Crucially they can also understand context so that you can have an informed conversation and answer subsequent questions more accurately.
Bing Chat Enterprise, which may be included with your M365 licensing is an AI-powered web chat with commercial data protection. For businesses concerned about the confidentiality of their company information, Bing Chat Enterprise offers a secure alternative to ChatGPT as any company data submitted doesn’t leave your tenant. By ensuring that data never leaves your tenant, Bing Chat Enterprise provides peace of mind.
Another example of generative AI in the pipeline is Microsoft’s highly anticipated Microsoft 365 Copilot. Due to launch for small and medium-sized businesses early next year, Copilot will be harnessed across the M365 suite, whether that’s Outlook, PowerPoint, Word or Excel. For example, in PowerPoint, you’ll be able to ask Microsoft Copilot to create a presentation based on a proposal document, set of notes and a file of images. Then you’ll be able to ask it to make your new presentation three slides shorter and Copilot will be able to do just that.
Azure opens the AI door for business
Open AI is not part of Microsoft, it’s a separate not-for-profit organisation and was responsible for the launch of ChatGPT. However, when you buy Open AI through Azure, you receive the added benefits of data privacy, residency, security, SLAs and integration with other Azure services. This means that you can deploy your own instances of large language models and securely build whatever systems your business needs using your own internal organisational data.
Azure Open AI and payment
Services like Microsoft 365 Copilot apply user-based licencing and are billed per user per month. However, with Azure Open AI, you pay for what you use on a token basis. With chatbots, when you ask a question, bothe the question and the reply you receive are represented in a number of words which are quantified as a number of tokens. A token is four characters on average, and your pricing is dependent on how many tokens go back and forth between the user and Azure Open AI.
Prompts enabling tailored experiences
In Azure Open AI you also refer to user and system prompts. A user prompt is the question you ask to the Azure Open AI service. A system prompt is how you configure the chat bot prior to deployment. These are important because they allow you to give your users customised experiences. For example, you can say that whenever a user initiates a response from the chat bot, it should greet them with a specific message. You can also programme it to limit your responses to a specific word count to keep your number of tokens and therefore your costs down. You can also use a technique called prompt engineering to educate users on how to ask questions which return more relevant answers.