What's New In Microsoft's AI Builder? (2024 Release Wave 2) [AI Builder Series: Article #4]
Microsoft's 2024 Release Wave 2 has been stacked with newly announced features, sparking well-received excitement within the Power Platform community. However, whilst the updates to Power Apps, Power Automate, and Microsoft Copilot were exciting, the AI Builder is again going under the radar.
Microsoft's list of new releases was by far the longest when discussing the AI Builder, and in this article, we will be looking at each one in detail so you can understand better what this could mean for you and your organisation.
AI Builder Announcements Summary:
Enhanced Governance & Administration (Public Preview Release: September 2024)
The Validation Station (Public Preview Release: March 2025)
Prompt Builder Changes (Public Preview Release: Dependant)
Enhanced Governance & Administration (Public Preview Release: September 2024)
I know, Governance can seem like a dull topic in the grand scheme of AI, but these features are quite important if you're planning to integrate the AI Builder into your current Power Platform infrastructure anytime soon.
Within "Governance and Administration" Microsoft have announced two key features:
AI Models can be managed through the Power Platform Admin Centre.
The AI Builder will become available in the Power Apps Developer Plan.
What can this mean for you?
AI Models can be managed through the Power Platform Admin Centre
Administrators will now have greater control over who can access and use the AI Builder offerings within different environments.
AI Models can be individually configured, with sharing and limiting access to different models being introduced.
The AI Builder will become available in the Power Apps Developer Plan
The Power Apps Developer plan is an initiative provided by Microsoft to encourage developers to experiment and test different Microsoft products within the Power Platform, free of charge. If you don't have an account yet and are serious about your Power Platform career, see how to create one by following the link below.
The AI Builder will now become an offering of the Plan, so users have free access to experiment with AI Builder capabilities. This would give you an easier way to test out the offerings without committing 100% to them and guide your decision-making to see if the software is right for you.
Currently, it's planned that Microsoft will offer you a free 20,000 AI Builder credits per month included in the plan (which is a good amount to have some fun with).
AI Builder Credits are the main pillar in the AI Builder's licensing model. If you want to learn more about how they work, please see the link below.
The Validation Station (Public Preview Release: March 2025)
This one's amazing. And again, under the radar.
Before I start explaining this feature, it would help to understand the following: "AI Prompts" = anything generated by the AI Model.
The Validation Station will be a new offering allowing users to provide feedback on AI Prompts used within Models.
If the AI Model created doesn't work properly or generates incorrect information, this can be flagged by the end user for the maker to review, notifying the maker so they can take steps to fix whatever went wrong.
Microsoft calls this a human-in-the-loop feature (their words, not mine) and states that it should achieve the following 4 aspects in its initial release:
User Feedback: Users can provide feedback on AI prompts when they aren't accurate.
Feedback Notifications: Makers are notified when new feedback is available.
Feedback Filtering: Makers can filter and list reviews sent for a given AI prompt.
Actionable Feedback: Makers can use the feedback received to increase the AI prompts' accuracy.
If you want to hear it in Microsoft's words see the link below.
Prompt Builder Changes (Public Preview Release: Dependant)
The Prompt Builder in the context of Microsoft's AI Builder offering is simply the interface where you build, test, and deploy AI Models (such as the document processing model).
There were 4 Prompt Builder changes/developments announced, which we'll go through one by one:
Few-shot learning and validation support.
Authoring Power Cards with GPT Prompts.
Processing images with GPT in the Prompt Builder.
Adding Unstructured Data Knowledge to GPT Prompts.
Few-shot Learning And Validation Support (Public Preview Release: March 2025)
First, let me explain what Few-shot learning is for those of you who may not have heard of it.
Few-shot learning helps builders develop machine learning models in a feasable manner, feasbale meaning they don't need crazy large data sets to do so, and can train models on smaller data sets.
Few-shot learning provides a range of benefits, from time-saving to accessibility for people who don't have access to big data. There can also be many cost-saving benefits associated with Few-shot learning, which allows makers the flexibility to develop more adaptive learning systems, as models will learn quickly.
If you want to learn more about Few-Shot Learning from an actual machine learning expert instead of myself, please refer to the article below.
We haven't seen this in action yet, so it's difficult to put our fingers on exactly how this would look. However, Microsoft promises that this feature will allow you to provide fewer specific examples of data inputs in the Prompt Builder to help create the correct prompt.
Authoring Power Cards With GPT Prompts (Public Preview Release: March 2025)
With the Prompt Builder, you'll be able to create something Microsoft is calling "power cards" that turn these prompts into dynamic visuals. This means you can generate accurate and interactive data displays with AI, making it easier to understand and present your data in the context of the Power Platform.
This feature makes it simple to convert AI-generated data into user-friendly visual components. You can use these power cards in different parts of the Power Platform, providing a more intuitive way to show AI outputs interactively.
Imagine you’re building an app and want to include AI-generated insights. Before, you might need to write complex code to turn AI data into visual elements. Now, with AI Builder’s Prompt Builder, you can write a natural language prompt, save it as a power card, and instantly get a reusable dynamic visual component that can be utilised in your app (I'm excited about this one!).
Processing Images With GPT In The Prompt Builder (Public Preview Release: March 2025)
The Prompt Builder will support image inputs with the GPT-4 with Vision (GPT-V) model. This means you can create solutions that understand and analyse images, unlocking new possibilities like understanding charts, infographics, scientific diagrams, and even solving math problems based on images.
Take the following example:
Imagine you're developing an app that helps students with their homework. With this new feature, you can create features where students take a picture of a math problem or a science diagram. The AI can then recognize the text and images, understand the context, and provide step-by-step solutions or explanations. For example, a student can take a picture of a chemistry chart, and the feature will not only identify the elements but also explain the chemical reactions and concepts displayed in the chart.
Just take a minute to think about the possibilities this could offer Canvas Apps!
Adding Unstructured Data Knowledge To GPT Prompts (Public Preview Release: November 2024)
Imagine supercharging AI Models within the context of your organisation through documents and websites. This feature will enable whole websites and documents to be uploaded and used in AI Builder prompts.
This new feature allows the model your training to understand the context of your business much easier, allowing for both documents and webpages to be used to train models and getting AI Models to understand the context of your organisation faster, in a more cost-efficient way.