A pragmatic approach to OpenAI usage

OpenAI, GPTs, and Bard have dominated tech news headlines in the past year, showcasing groundbreaking use cases and seemingly futuristic trends. It is pretty evident with the number of ChatGPT downloads since its first release and everyone, from your tech-savant friend to your next-door neighbor, thinks that we are now living in an era of technological marvels.

However, the reality of how AI can truly transform businesses is a more grounded perspective. Natural Language Processing like text recognition or OCR for more than a decade. Products like Kofax have been able to read invoices long before AI gained widespread attention. While AI has made strides in tasks like generating, translating, and explaining programming logic, it primarily serves as an assistive tool rather than a fully autonomous coding machine. In addition, analysis shows that 52% of ChatGPT answers from Stack Overflow questions were incorrect highlighting the need for human analysis to determine their accuracy. Story generators and essay-writing bots, while entertaining, have limited practical utility.

What people think AI is

What people often envision AI to be is a transformative force that can single-handedly solve world hunger or replace human jobs overnight. However, the reality is more nuanced.

What AI really is

In practice, AI's greatness lies in its ability to augment existing processes with simplicity and efficiency. At my current organization, SoftwareOne, we've harnessed Microsoft's Azure OpenAI to create a practical solution with minimal effort, cost, and time investment. We've developed a straightforward MS Teams helper bot that handles routine user tasks like password resets. One of its key functions is a Question and Answer feature integrated with Azure OpenAI. While it may not address grand challenges like world hunger, it offers valuable assistance by quickly providing meaningful information. We maintain our knowledge base within a leading corporate SaaS and treat it as our source of truth for operational and how-to information. Given that the data is stored in SaaS, we have limited flexibility in working with it. Adapters available in the market can provide more extensive capabilities but often come at a premium cost.

Simple Use Case: Question and Answer via MS Teams

To illustrate a simple use case, we've devised a seamless OpenAI conversation flow that leverages trusted data not publicly modeled or extracted and is accessible by an application that everyone in the organization uses, MS Teams. We extract changes from our corporate wiki using a Serverless Function App that runs daily, storing this information as PDF files in a Blob Storage, which is then indexed by our Search Service. OpenAI utilizes the data model from this Search Service to deliver results when a user sends an "Ask KMDB" query through our Bot Service within our organization's MS Teams environment.

As a result, we have created a practical and efficient usage of Azure OpenAI that provides precise and reliable data from our Corporate Wiki supported by references and citations with links to our Wiki page. Importantly, it doesn't generate fabricated responses based on publicly available sources. The bot is easily accessible via an application that everyone uses daily and can be invoked with a few keywords powered by Azure Conversational Language Understanding. The time to deployment is relatively quick and the operating costs are minimal. It simplifies access to wiki knowledge, eliminating the need to visit the wiki site and search for information, while also promoting greater utilization of the wiki.

In conclusion, AI is a powerful tool that shines brightest when integrated with other tools and processes. It's not the fantastical, world-changing technology that often garners headlines but is left unimplemented due to overly ambitious business plans. While AI's potential is vast and theoretical use cases are impressive, it's prudent for businesses to explore practical, present-day applications. We may not yet have fully automated teams of AI bots creating production-ready applications (ie. ChatDev) or AI agents conducting complex business operations, but we can certainly harness the capabilities at our disposal and continue to progress.