How Generative AI will improve digital messaging, no matter what happens with ChatGPT
Before we jump into Generative Artificial intelligence (AI) and ChatGPT, it’s important to remember that they exist within the broader context of AI. Artificial Intelligence has been discussed since ancient times and computer-driven AI has been diligently worked on since the advent of the electronic computer. In the last few decades, we’ve seen AI-driven algorithms defeat our greatest chess players and even detect cancer with greater accuracy and speed than trained doctors. But until recently, far less attention had been paid to Generative AI—the ability for computers to learn the structure and syntax of a language, rather than just its meaning or content, and use it to create something entirely new.
When ChatGPT launched as a free tool in November of 2022, it took the world by storm. Within days they had over 1 million users, and it seemed to many that ChatGPT had come out of nowhere. The team behind the technology was shocked by its popularity, in part because the technology powering it isn’t new.
What is Generative AI
In the most simple terms, Generative AI refers, broadly, to any type of AI that can be used to generate new content. An example would be Large Language Models (LLMs) which are able to create new combinations of text in a way that sounds natural. Other examples of Generative AI are deep generative models and reinforcement learning agents.
Large Language Models (LLMs) or Foundational Models are a vitally important tool in the field of Generative AI.
They are trained on massive datasets of text and code, which helps them understand the statistical relationships between words and phrases. While LLMs can craft content and responses to questions that feel broadly ‘human’ it’s important to remember that many are still under development. Like the humans that they emulate, they can sometimes generate text that is offensive and/or factually incorrect.
For many business use cases, figuring out when, where, and how to leverage Generative AI is the key to using it in a way that adds value. One example would be in a chatbot experience, tapping into a tool like ChatGPT to help an automated bot better understand multi-intent prompts from users.
What is ChatGPT
At a high level, ChatGPT is a type of Generative AI model that has been trained using human feedback to provide responses that people are more likely to enjoy and find value from. The base of the human feedback training came from letting real people use the tool and provide feedback. ChatGPT was developed by OpenAI and leverages earlier foundational LLMs.
ChatGPT was released for free and in a conversational interface that was accessible and easy to use. And oh how it was used!
Building a ChatGPT integration into a digital messaging platform does not, by default, build in added value for the user or business.
Given the broad user appeal of ChatGPT, it can be easy to see them as the Generative AI solution. To do that would be a mistake because the technology is so much bigger than one application.
The Continued Importance of Foundational Large Language Models (LLMs)
If you look at how Generative AI models are built and trained, you quickly start to understand that they are only as good as the data that is fed to them. By selecting what they ingest, you can shape their worldview and all future outputs.
With Generative AI tools that range from supportive AI companions to Google Bard, which pulls information live from the web, companies today have the ability to choose very specific tools that best meet their very specific needs. When you start to dig in, it seems unlikely that every customer of every Conversational AI platform would have their needs best met by one single platform. What is infinitely more likely, and what we’re betting on at XAPP AI, is that the continued innovation in the Generative AI space is going to mean that what works for one company might not work for another.
Rather than build our platform around a single tool like ChatGPT, we’re building it to work seamlessly with many Generative AI tools.
The Importance of Extensible Architecture
As a platform serving the needs of a broad array of customers, we see Generative AI and LLMs as constantly evolving technologies that require architectural flexibility.
What that means, functionally, is that we’re built in a way that allows for flexibility when selecting a Generative AI tool to help support automated, natural, and informational conversations through digital messaging mediums.
When conversations are initiated and questions are asked, we leverage each client’s specific Unified Knowledge Model™ before looking more broadly at their industry model (which is trained on data that helps the AI differentiate between things like what “tile” means to a roofer versus a plumber.) If the answer can’t be found, we then have the ability to tap into a Generative AI tool. Our platform’s extensible architecture allows us to select the best in breed Generative AI for each specific use case, without requiring an entirely new build of our platform.
If your business needs the ability to search the web in real-time for a customer question tied to current events, you likely would not choose ChatGPT, which is restricted to using historical data. A tool like Google Bard, which has the ability to leverage the broader web to find information on current events, would be a better fit.
We’re constantly training our models in an ongoing learning loop, gathering information, analyzing it, and using it to improve automated digital messaging performance.
That iterative and ongoing process includes deep dives into the further refinement of our model based on broader informational inputs from Generative AI that can offer up more contextual and current information, more efficient lead intake, routing, and scoring, and an overall better user experience.
While many in the industry have quickly stood up products that leverage ChatGPT to ride the wave of press, we continue to view Generative AI technology more broadly as a tool that can be used to strengthen the learning loop for our customers and improve digital messaging.
When building a platform, you want to look at longevity, stability, and scalability. A digital messaging solution that anticipates different businesses and industries having different informational needs, should plan to meet them. And so we have.