ChatGPT, the research chatbot released for public use by OpenAI has reignited discussions on the power of conversational AI and left many CX leaders wondering what impact it will have on customer service. Will it replace agents? Has it made everything innovation-first companies have invested in conversational AI over the past few years obsolete?
Our VP of AI Technology, Yves Normandin, shares his thoughts on the influence ChatGPT will have within the contact center.
What is ChatGPT?
ChatGPT is a large language model (LLM) from OpenAI. Major tech companies like Google, Meta, and more are constantly competing with one another to create increasingly complex and powerful LLMs. In short, an LLM is a deep learning model, trained on large amounts of text, that simply predicts the next word in a text given previous words. Using the model to generate one word after another, it produces sentences, paragraphs, and even texts that are surprisingly well-constructed and relevant.
How big is big?
A model like ChatGPT has over 170 billion parameters and was trained on over 10 GB of text.
So, it knows everything about everything then?
ChatGPT does have its limitations. The model is strictly based on its trained parameters and the previous words in the text or conversation. It doesn’t – and can’t – access any outside information or database to make its predictions. ChatGPT knows nothing about anything that happened after 2021.
What makes ChatGPT so special?
What makes ChatGPT so special is that it is specifically optimized for conversations. Contrary to previous LLMs, ChatGPT remembers information exchanged in past interactions so you can ask follow-up questions, challenge incorrect answers, etc. The result is mind-blowing! Another reason why it’s getting so much attention is that OpenAI is available for anyone to try, and people have been sharing incredible conversations they’ve had with it.
Does this make everything companies do within the contact center with conversational AI obsolete?
The short answer is no. As amazing as the technology is, it’s not quite clear yet how it can be used to help create effective virtual agents. The technology still has some important limitations, for example:
- The model only knows the data it was trained on, so it cannot easily leverage external information like customer data- a must for any agent, virtual or human.
- And we have very little control over what the model generates. It will often produce plausible-sounding but incorrect or nonsensical answers. This is what we call “hallucinations.” For customer service, that’s just not acceptable.
In fact, Sam Altman, CEO of OpenAI, tweeted, “…it’s a mistake to be relying on it for anything important right now. it’s a preview of progress; we have lots of work to do on robustness and truthfulness.”
So far, the best use cases for LLMs have been those where a human can validate the generated text or use it as a suggestion to be curated. We can already see examples of this in autocomplete features for text or email messages or to suggest chat responses to customer service agents. Powerful autocomplete features are also available for writing software such as GitHub Copilot. As with Jasper AI, LLMs are also used to help create content or copy.
That’s not to say we can simply ignore LLMs. Big companies are investing millions of dollars into the technology, so it’s evolving rapidly- we need to pay very close attention to ongoing developments. But for now, businesses must continue to leverage proven technologies that can be easily trained on all sorts of relevant business and customer data to create the best possible virtual agents and experience for their customers.
Just for fun, we asked ChatGPT what the best way would be to use ChatGPT to create customer service virtual agents. Here’s what it responded:
ChatGPT is a neural network-based chatbot model that uses the GPT-3 (Generative Pretrained Transformer) architecture to generate human-like responses to user input. To use ChatGPT to create customer service virtual agents, you would first need to train the model on a large dataset of customer service conversations, such as email or chat transcripts. This would allow the model to learn the language and common questions and answers used in customer service interactions.
Once the model has been trained, you can use it to generate responses to user input in real time. To do this, you would need to integrate the model into a chat platform or other application that allows users to interact with it. You could also use natural language processing (NLP) techniques to improve the model’s ability to understand user input and generate relevant responses.
Overall, using ChatGPT to create customer service virtual agents can be an effective way to automate and improve the quality of customer service interactions. However, it’s important to remember that the model’s responses will only be as good as the training data it was trained on, so it’s crucial to use high-quality, relevant data to train the model.
