The Evolution of AI: From OpenAI's GPTs to Microsoft's Copilot Studio

In our journey through the rapidly evolving landscape of AI, we've witnessed remarkable strides from OpenAI's easy-to-use GPTs to the sophisticated integration of AI in Microsoft's Copilot Studio. Today we will explore the recent progress since the November 2023 OpenAI Developer Day Conference and its implications for the future.

OpenAI's GPTs: Democratizing AI Development

OpenAI revolutionized the AI space with GPTs (Assistants) within ChatGPT. These are highly customizable versions of ChatGPT, allowing users to create AI models tailored for specific tasks or topics. The beauty of GPTs lies in their simplicity and accessibility:

• User-Friendly Creation Process: No coding skills are required to create GPTs, making it accessible for experts in any field. They are designed to be user-friendly, allowing individuals without coding experience to create and modify these models. This is achieved through an intuitive interface where users can define the behavior and capabilities of their GPT by simply conversing with the GPT Builder.
• Integration of Various Tools: GPTs in ChatGPT can integrate various tools like web browsing, image creation through DALL·E, and data analysis, broadening their potential use cases.GPT Store: OpenAI recently launched a GPT Store, where users can feature and monetize their GPTs, providing a simple AI App Store for users to discover and consume new capabilities.
• Diverse Applications: GPTs cater to a wide range of applications. For instance, a GPT could be created to assist with language learning, where it is fed specific language rules and examples, or it could be developed to offer technical support, equipped with detailed technical documentation and troubleshooting guides.

OpenAI's GPTs are designed to be customizable and task specific. This approach suggests a shift towards more specialized AI applications, where individual bots are tuned for particular functions, offering more precision and efficiency in those areas vs a single bot such as Siri or Alexa.

Microsoft's Copilot Studio: Elevating AI with Advanced Capabilities

Microsoft's integration of Generative AI into Copilot Studio signifies a leap in AI applications, especially for businesses and developers. Here's how Generative Answers and Generative Actions enhance the Copilot Studio experience:

Generative Answers

Generative answers in Microsoft Copilot Studio allow your custom Copilot to generate responses based on a variety of sources. This feature is crucial in expanding the conversational reach and relevance of AI agents.

• Source Diversity: Generative answers can utilize various sources, including external resources like Bing Web Search and Bing Custom Search, as well as internal resources such as Azure AI Search, SharePoint, and documents uploaded to Dataverse. The SharePoint data source seems to be powered by Microsoft Search today but I expect this to improve with Sematic Index for Copilot.
• Custom Data Sources: If the necessary data does not exist in a supported source, you can provide your own data through Power Automate Flows. This data is formatted into a JSON object and used to generate answers. The custom data field accepts a JSON array of objects, each representing a ContentLocation/Content pair. The copilot generates answers from the Content and includes a link to the data source in ContentLocation.
• Search and Summarize Capabilities: Within any copilot topic, you can add a 'Create generative answers' node. This node allows for specifying additional sources that the node searches based on your inputs. These sources can override the ones specified at the copilot level which function as a fallback of a question isn't answered in a specific topic.
• Operational Factors and Settings: For effective use of generative answers, it's crucial to choose a source that the AI system can easily search and summarize. The capability is designed to query knowledge from the selected website and package relevant findings into a concise response. Always test and review your bots before publishing and consider collecting feedback from bot users.

Generative answers in Microsoft Copilot Studio represent a significant step in enhancing AI-assisted conversations, providing a more dynamic and context-aware response mechanism. By leveraging various data sources and continuously refining based on user feedback, these AI agents can offer more accurate and relevant information in a conversational format.

Generative Actions

Generative Actions: This feature allows copilots to respond to user queries by selecting appropriate topics or plugin actions. The selection is based on the descriptions associated with each topic or action, with the system using GPT to determine the best match for the user's query.

• Dynamic Chaining: Dynamic chaining enables the copilot to trigger multiple topics or plugin actions sequentially, based on the user's query. This allows for more detailed and comprehensive responses, as the copilot can address multiple aspects of a query in an ordered and coherent manner. For example, if a user asks about a store's location and its hours, the copilot can first provide the address and then the store hours, effectively covering both aspects of the query. This new composability on-the-fly will dramatically change how we think about developing topics with smaller more granular outputs designed for run-time chaining.
• Plugin Actions: These are actions based on prebuilt or custom connectors and Power Automate cloud flows. Plugin actions can generate contextual responses to user queries. They can be used to query other systems and provide detailed answers. For example, plugin actions can be configured to send emails, post messages in Teams, or get weather forecasts, among other functions.
• Customization and Ease of Use: With dynamic chaining and generative actions, there is less need for manually entering different trigger phrases for each topic, as a clear and natural description of what each topic does is sufficient. This simplifies the task of building copilots and makes it more accessible.
• Integration with Power Automate: Dynamic chaining and plugin actions can be integrated with Power Automate, allowing for even more complex and useful actions, such as managing HTTP requests or returning data in a specific format.

Dynamic chaining and generative actions in Microsoft Copilot Studio significantly enhance the capabilities of AI agents, making them more intuitive, responsive, and useful for a variety of applications. These features represent a substantial improvement in the field of conversational AI, offering more dynamic and flexible solutions for interacting with users.

What's Next - The Future of AI in Business and Development

At the Microsoft Ignite Conference in November, a week after OpenAI announced GPTs, Microsoft announced that GPTs would be coming to Copilot Studio. There were no details shared other than a Figma mockup of the navigation showing GPTs as a first-class capability.

Imagining the integration of GPTs into Microsoft Copilot Studio opens up a myriad of possibilities, especially considering the diverse data sources and tools like Power Automate workflows. No details have been announced from Microsoft yet, but we know enough about both GPTs and Copilot Studio to imagine the possibilities.

1. Enhanced Customization with GPTs: Integrating GPTs into Copilot Studio could allow for the creation of highly specialized AI agents tailored to specific tasks or industries. GPTs could be published as specific Topics, being more adaptable and customizable as simple AI applications rather than simple conversation flow. These GPTs could come with unique knowledge bases, making them more effective in niche areas or specific business functions.
2. Diverse Data Sources for Informed Responses: GPTs in Copilot Studio could leverage a range of data sources, similar to the generative answers feature. This might include Azure AI Search SharePoint, OneDrive, and Dataverse, GPTs could access and incorporate a company's internal documents, reports, and data sets for more informed and contextually relevant responses.
3. Public Web Data Grounding: Integrating Bing Web Search or Custom Search as data sources, GPTs could pull in current, publicly available information to enhance their responses with up-to-date data. This is a huge gap in Copilot Studio today as Generative Answers is limited to a couple websites defined by the bot author.
4. Power Automate Workflows as Tools: Power Automate could serve as a powerful tool in conjunction with GPTs, enabling automation and interaction with a variety of systems. For instance:Automating Routine Tasks: GPTs could trigger Power Automate workflows to perform routine tasks like data entry, scheduling, or even initiating complex business processes based on the user's interaction with the Copilot.
5. Dynamic Data Retrieval and Processing: GPTs could use Power Automate to retrieve data from various sources, process it, and use it to inform responses or decisions. This could be particularly useful in scenarios like customer support, where real-time data retrieval is essential.
6. Custom Integrations and Extensions: With the ability to connect to custom APIs and services, GPTs in Copilot Studio could extend their capabilities beyond standard offerings. This might include integration with CRM systems, analytics tools, or even industry-specific software.
7. Contextual Understanding and Response Generation: Leveraging the advanced NLP capabilities of GPTs, the AI agents in Copilot Studio could provide more nuanced and context-aware responses, understanding user intent more accurately and offering solutions or information that is highly relevant.
8. Continuous Learning and Adaptation: With the integration of GPTs, these AI agents could continuously learn and adapt based on user interactions, feedback, and changing data, ensuring that their responses and actions remain relevant and effective over time.
9. Enhanced User Experience: The integration could lead to more intuitive and conversational interfaces, making it easier for users to interact with AI agents and get the information or assistance they need in a natural, conversational manner.

The potential integration of GPTs into Copilot Studio hints at a future where AI agents are not only more intelligent and responsive but also deeply integrated with a business's data and processes, offering tailored, efficient, and highly effective solutions.

As we continue to navigate these exciting developments, it's clear that the potential for AI to transform industries and enhance our capabilities is immense. Whether you're a seasoned developer, an AI enthusiast, or a business leader, the future of AI holds limitless possibilities.