The launch of chatGPT has sparked a lot of talk about AI in recent months, whether in the news or everyday conversations.
Chatbots and self-care tools have been at the heart of the customer relationship for several years now:
- 96% of consumers have heard about chatbots and know what they are
- 62% of consumers prefer to find answers themselves
- 40% of savings on customer support costs with bots
And this trend could well accelerate with the adoption of GPT!
But before we go any further, here’s a quick disclaimer…
GPT is constantly evolving. This article may be out of date tomorrow. But the work we’re doing at Dydu to integrate GPT into our solution is still ongoing. So, stay tuned!
ChatGPT: Why all the hype?
There’s nothing new about chatbots, so why is there so much buzz around Open AI’s ChatGPT?
If you’ve ever had a conversation with chatGPT, you were probably impressed by:
- the feeling of being understood by a chatbot and the very natural nature of the conversation
- its ability to write poems, press articles, or even legal documents with precision and/or creativity
- the bot’s quick answers
The bot can express itself perfectly, whether in English or French.
ChatGPT also has the ability to remember past conversations and provide contextualized answers.
In other words, it’s (almost) like talking to your best friend! ChatGPT has positively changed people’s perception of chatbots. Which is great news!
Example of a conversation with chatGPT:
Can you explain the benefits of a chatbot as if I was 10-years old?
Of course! A chatbot is a computer program designed to talk to people in a friendly manner. You can ask questions to which it will try to reply. Here are a few advantages to using a chatbot:
- Speed: chatbots can answer your questions instantly.
- Availability: you can talk to a chatbot any time of day or night, even when nobody else is available.
- Non-judgmental: you can ask a chatbot anything you want without worrying that it will judge or poke fun at you.
- Alert: unlike humans, chatbots never get tired of answering questions.
- Fun: some chatbots are designed to be funny or to tell jokes.
In other words, chatbots can provide quick and reliable answers, while being available all the time and entertaining.
What are the differences between chatGPT and Dydu’s chatbot solutions?
Before we talk about the differences between chatGPT and Dydu, let’s take a look at the similarities. Both specialize in natural language processing (NLP):
ChatGPT uses its own language model, GPT.
- Dydu also owns its own NLP component, which is at the heart of its chatbot solutions
But the two solutions don’t offer the same user experience.
ChatGPT and Dydu take a different approach to processing language and learning.
ChatGPT uses unsupervised learning, i.e., it learns from its interactions with users. It is non-deterministic (it can give different answers to the same question) and non-explainable.
It is risky for brands to use because you can’t control what it says. ChatGPT could make false claims or be racist, sexist, etc., and seriously damage a brand’s image or create a bad buzz.
Example of a false answer provided by chatGPT :
How many calories are there in a horse egg?
Humans don’t often consume horse eggs, so there is very little nutritional data available. However, generally speaking, horse eggs have a similar nutritional value to chicken eggs in terms of protein, fat, and calories.
There are, on average, approximately 70-80 calories, 6 grams of protein, and 5 grams of fat in a chicken egg. Horse eggs may have a similar nutritional value, but it depends on the size of the egg.
It is worth noting that eating horse eggs is not common practice in most cultures and could even be considered unsafe. If you’re unsure whether it’s safe or not to eat horse eggs, make sure to speak to a health professional or nutritionist before including them in your diet.
Dydu has opted for supervised learning. You can use our back-office to choose whether or not to add new knowledge articles and to manage the bot’s answers. This gives companies total control over what they say to their customers.
Bot management via a back office
Unlike Dydu, ChatGPT does not offer the option of managing the chatbot via a back-office.
The Dydu back office is one of our solution’s much-appreciated primary assets. It is intuitive and enables to:
- view dialog history
- analyze the relevance of the bot’s answers and customer satisfaction
- enrich the knowledge base and therefore manage the content
- check use and dialog analytics
None of these features exist in Open AI’s product. Our tool and solutions are much better suited to company needs for customers or employees.
Interconnections with tools and platforms
Unlike chatGPT, our software can interconnect with the main tools on the market, integrate easily with third-party channels or apps, and perform mass processing. It can retrieve data from your CRM, HRIS, ticketing tools, and other databases to personalize its answers.
We are developing more and more interconnections, in line with our clients’ needs, to:
- easily deploy our chatbots where users are (Microsoft Teams, Facebook Messenger, Instagram, etc.)
- improve the user experience by providing personalized answers and performing actions directly within the bot through company tool integration
Unlike chatGPT, our Dydu solution is GDPR compliant. Data security and GDPR compliance is extremely important for a solution like Dydu with CAC40 clients in the banking, telecoms, public sector, etc.
We host our data in France in an encrypted manner. We know where our data goes and how it is used. The way ChatGPT uses data is currently the subject of debate in Europe, prompting the CNIL to open an investigation.
How Dydu is integrating GPT into its solutions
We’re currently working on several POCs with some of our clients to leverage GPT and other language models, such as Meta’s LLaMA, within the Dydu engine.
The combination of both solutions could further improve the customer experience.
Integrating language models such as GPT with the Dydu solution is ideal for answering less frequent questions – i.e., those not covered by the knowledge base.
How? By using GPT with a pre-existing document base. If the Dydu bot can’t find an answer in its knowledge base, it can search the document base and answer more questions.
Accelerated Dydu bot training
Another interesting area to explore is automatically enriching the matching phrases in the knowledge base with GPT’s generative capabilities.
If you provide one matching phrase for each knowledge article, GPT can offer alternatives and speed up our engine’s learning process. All you need to do is check that the phrases suggested by GPT are relevant.
On 12 April, the Dydu co-founders discussed all these points in more detail in a ChatGPT webinar.
[ Only in french ]