A LOT of information has been swirling ever since ChatGPT hit the news!
ChatGPT! Everyone uses it, right? Fastest growing application of all time! ChatGPT is an AI model that uses deep learning to generate human-like text based on prompts from users like you. It works by predicting the next word in a given text, based on the patterns it has learned from a massive amount of data during its training process. Here’s an example:
Well, what happened since then? In the past 6 months, Microsoft Ignite showcased work on Microsoft products integrating some of these Large Learning Models into Microsoft products.
GPT means Generative Pre-Trained in terms of the AI model. These are Large Learning Models of significant size, generally trained on “much of the internet” for their information contained within.
OpenAI has the mindshare in recent history on these GPT models. There are others, like Amazon’s Bedrock, and Google’s Bard and Salesforce’s Einstein, and Meta’s Llama
Transparency note: I work for Microsoft at the time of this writing.
Microsoft is a 49% investor in OpenAI, and besides investing in OpenAI, also hosts these (and other) models. The Microsoft version of OpenAI are called Azure OpenAI on the Azure cloud. These models on the Azure cloud are bound by the Responsible AI principles. That topic is huge, and in general covers using AI for good, and making sure your data is yours, and your deployment of your model in your Azure tenant is not trained on your data.
These OpenAI Models are being leveraged inside of Microsoft’s product suites to build per-product “Copilots” relevant to those use cases. Why “copilot”? Because you, the human, are still the pilot.
Work Productivity has been shown to increase when leveraging some of these tools within the Microsoft 365 suite of products as well as other Microsoft products. “CoPilot for Microsoft 365” is a set of copilots built for different use cases inside the broad Microsoft 365 experience. Here is a quick video on M365 Copilot. Here is a 56 minute video on a variety of copilot experiences, from office, to sales, to custom copilots.
Think of it this way: Copilots are a way to offload the long tedious work, say to a smart, but inexperienced, human assistant.
“Go build me a powerpoint on Gross Domestic Product for the last 20 years” is an example of a human command. Give that to your new hire or intern, and in a day they come back with something. I used the above prompt in PowerPoint copilot to have it generate a draft, including data, of the request. It took about 30 seconds max. I re-did the prompt, and it actually presented a different view of that information into a different set of slides. Here is an example:
Inside of Teams, for example, I installed the M365 Chat app. Here is what it says about itself:
M365 Chat
The Microsoft 365 Chat feature in Copilot combines the power of large language models (LLMs) with the intelligence of the Microsoft Graph to help you get things done. If authorized by your organization, you have the option to add plugins for more data sources (like apps and web content) that Copilot will interact with on your behalf. Copilot can synthesize data from multiple sources to give you a summary of things you need to catch up on, including your files, messages, meetings, emails, and people. It can also help you find and use info that’s buried in documents or lost in conversations. And with Copilot by your side, you can create content with it all. Ask a work-related question, or try one of these: – Draft a message with action items from my last meeting – Catch up on my unread email – How do I write a request for proposal? AI-generated content may be incorrect, so sources are provided for your review when possible. Discover more with these helpful links: – [Explore what’s possible with Copilot](https://go.microsoft.com/fwlink/?linkid=2240275) – [See what’s new with Copilot](https://go.microsoft.com/fwlink/?linkid=2239566) – Check out our [FAQ](https://go.microsoft.com/fwlink/?linkid=2238505)Created by Microsoft Corporation
Here’s what this app can do:
•Receive messages and data that I provide to it.
•Send me messages and notifications.
•Access my profile information such as my name, email address, company name, and preferred language.
•Receive messages and data that team members provide to it in a channel.
•Send messages and notifications in a channel.
•Access this team’s information such as team name, channel list and roster (including team member’s names and email addresses) – and use this to contact them.
In M365 chat, I can say “Summarize recent emails from coworker” and it will show many in summary form. What a time saver! Here is an example prompt and result. Note: It always gives you links to the source, so you can validate the accuracy. This also works in the mobile version of teams with the M365 chat app installed.
That’s pretty good. But have you ever wondered where the approved customer-ready powerpoints are? This is where M365, data labelling, and other features shine, with security. In the following case, I will not show the results because they are Microsoft internal. It did return 5 examples.
There are licensing and other requirements for Microsoft 365 Copilot.
Copilots are not just in Microsoft 365. Copilots for Power BI, the new Microsoft Fabric data and ai platform, and more will have prebuilt copilot experiences.
Using the AI-infused Copilot in Dynamics 365 Sales, Avanade reduces workload for sellers with a range of time-saving features.
Keep in mind: These Large Language Models have limitations on accuracy. They are “Creative” and there are settings on “how creative” they should be. You are in charge. You cannot just “copilot and send”. The above m365 chat example which is new, may not always do things in the right order or precision by default. Additional prompt engineering can help with accuracy of results.
For developers who wish to build “Copilots” into their own products, or ISVs who wish to improve their applications similarly can use Azure OpenAI and other Large Language Model-as-a-Service (announced at Microsoft Build in May 2023). You can integrate the latest AI models, such as Llama 2 from Meta and upcoming premium models from Mistral, and Jais from G42, as API endpoints to your applications.
As with other AI platform infrastructure on Azure, you do not need to manage the GPU setup, hardware, patching, or other processes.
To help you train or use or implement the variety of AI, including the above mentioned models, there is Microsoft AI Studio. To Quote: “Your platform for developing generative AI solutions and custom copilots”
Finally for a low code alternative, there is Microsoft Copilot Studio. “Copilot Studio offers graphical development environment to build copilots using generative AI, sophisticated dialog creation, plugin capabilities, process automation, and built-in analytics that work with Microsoft conversational AI tools.”
I encourage you to “Try a demo” in the above website, enter your favorite website with information on it, and ask questions of that. My example would be: http://www.microsoft.com and ask who is the ceo, or president of the board for example. Check this out: My question is on the right.
I hope this information helps, and you can find value in a new way to accelerate your work.
Here is the link to the Ignite Book of news.
After writing most of this blog, I found the “AI Assistant” in WordPress. It gave me tips to improve the content before publishing! I want your input however.
Please let me know if you have any questions, and I encourage feedback!
You must be logged in to post a comment.