Website Chatbots: The new generation

By Andy Mundell | Published: 03 March, 2024

You've likely noticed the surge in customer-service chatbots popping up on popular websites you visit. Increasingly, having 'real-time' customer service and support is playing a pivotal role in sales conversion and business success. So, now's a good time to understand some fundamentals. Because not all chatbots are created equal.

Chances are, any experiences you've had to date with website chatbots have been interactions with "rules-based, Natural Language Processing (NLP)" systems". NLP systems have dominated the chatbot landscape for a number of years. However, a seismic shift occurred in March 2023 when OpenAI launched their public APIs, allowing 3rd party developers to leverage ChatGPT's large language models. This changed "everything!"

Before continuing it's important to point out here that there's a wide range of NLP based chatbot solution out there that vary greatly in their abilities. This article seeks only to address the key difference between NLP and GPT based technology. Not the individual chatbot solutions themselves.

When we talk about NLP (Natural Language Processing) chatbots versus GPT (Generative Pre-trained Transformer) chatbots for websites, we're essentially looking at two generations of technology that power how bots understand and respond to user input.

NLP Chatbots:

Basics: These chatbots are designed to understand human language through a set of predefined rules and patterns. They often rely on keyword matching and specific triggers to generate responses.
Strengths : They're good for FAQs and common, straightforward tasks where the queries can be anticipated and mapped out in advance. This makes them highly reliable and predictable for specific interactions and well suited for categorising and funneling the user experience
Limitations: NLP chatbots can struggle with complex or nuanced queries as they rely heavily on predefined patterns and might not understand variations in how people phrase their questions. Their responses can feel less natural and may fail to grasp the context or intent behind a user's question. Their abilities rely heavily on the quantity and quality of data and the pre-defined relationships and parameters encoded. Highly capable NLP systems require hundreds of thousands, or even millions of parameters. As a result they can be time consuming and expensive to build, fine-tune and maintain.

GPT Chatbots:

Basics: GPT chatbots, like Cleva.Bot, use "Artificial Intelligence - Generative Pre-trained Transformers (GPT)" pre-trained on vast amounts of data encompassing Trillions (yes, with a "T") of parameters. This training allows them to generate human-like responses that are logically reasoned and contextually relevant.
Strengths: GPT chatbots excel in understanding and generating natural language. They can handle a wide variety of queries, even those not explicitly anticipated, and are proficient at reasoning answers based on accumulative knowledge. They're great at maintaining context over a conversation, providing a better user experience and making interactions more engaging. They're also easier to build and maintain, meaning a lower cost of entry.
Limitations: While significantly more versatile, GPT chatbots can sometimes generate unexpected or incorrect answers because their responses are generated from relative logic in data, not pre-defined, hard-coded values. As a result, a period of chatbot training is necessary during deployment to identify and address any inaccuracies.

In essence, the leap from traditional NLP chatbots to GPT-based ones is analogous to moving from a manual to automatic transmission, where the machine takes over most of the heavy lifting.

Instead of relying on a fixed database of responses, GPT chatbots provide a more intuitive user experience. Generating unique responses in real-time and in context with the user's specific needs.

If you're still left struggling to understand exactly what a GPT powered chatbot actually is, then just try to imagine if "Clippy" and "Mr Data" had a lovechild, and you'll be getting pretty close.

Cheers..
&E