What are LLMs like GPT and PaLM?
OpenAI’s ChatGPT has created huge awareness around text-based generative AI. It doesn’t have a conscious understanding of what it’s writing like we do, but produces coherent text by simply predicting what the next word in a sequence might be. To do this, ChatGPT and other text generators use a type of machine learning called a large language model (LLM).
The LLM that supports ChatGPT is called GPT (Generative Pre-trained Transformer) and OpenAI release new versions of their model over time, from GPT-2 to GPT-3.5, then GPT-4.
Other notable LLMs include PaLM from Google, LLaMA from Meta, and Claude from Anthropic, but new models and enhancements are emerging all the time.
The most advanced AI assistants already use natural language processing (NLP) to recognise what your customers are saying. Now, with the integration of LLMs, we can significantly expand those capabilities to transform your customer care.
Launch a customer service chatbot in 10 minutes using LLMs
AI assistants like ours are already in the process of displacing traditional chatbots because of their speed, low cost and accuracy, and using LLMs, your AI assistant can now go live in less than 10 minutes using only your website URL to set it up. In as little as two minutes, your AI assistant is now ready to become an instant expert on your brand, business, and customer needs.
Turn your web content into a fully-functioning AI assistant
LLMs help us pull in information from your website and use it to automatically produce a list of question-and-answer sequences for your AI assistant to work from (we call these flows). This means all the content you’ve already approved for talking about your products, services and offers can be used to instantly create an AI assistant that knows how to answer customer questions. You don’t need to do anything but check you’re happy with the wording of each flow, or if you’d rather not use it, simply unpublish it.
When your AI assistant starts replying directly to customer questions, the response for each topic stays the same every time, so customers can come to rely on accurate, up-to-date responses whenever they need help, and they’ll get it instantly.
Get expert suggestions to add, update or develop your flows
Once you go live, our expert team of language experts, psychologists and writers review conversations customers are having with your AI assistant to make sure it’s always working at peak performance ― PII (personal information) is automatically removed. Working behind the scenes as your human in the loop, we send you suggestions for flows you can add to train your AI assistant on new topics of conversation with your customers, making sure it’s always up to speed with what they want to know. All you need to do is accept or reject the training suggestions and nothing goes into a flow without your approval.
Manual updates
If you want to add a new flow of your own, LLMs can help here too. To train your AI assistant manually, you’ll want to give it a few different examples of how a question might be phrased by different people. “I want to buy a t-shirt,” “what t-shirts are available,” and “have you got any T-shirts on sale” are all effectively the same question. If you can give your AI assistant a few different examples of phrasing to learn from, it’ll soon pick up the link. Within our platform, you can use LLMs to generate alternate phrasing ideas instantly. An LLM like GPT-4 has absorbed so much information from so many different places, it’ll never be short of suggestions to keep you moving through the process. Just accept the suggested phrases you want to use.
Hit the highest success rates faster with LLM support
The average success rate for the advanced AI assistants we create is 91%. If your AI assistant can recognise a request from your customer, understood what they need, and given a helpful response without any input from your customer contact team, that’s a successful interaction. Competitor platforms like Intercom have used GPT-4 to achieve a 50% success rate, but a 99% success rate is achievable. Bearing in mind around 1% of questions will always be either irrelevant (like someone asking a travel company what the football score is) or ones the AI assistant isn’t yet trained to answer, so there is no 100%.
Now we’re using LLMs to create AI assistants in minutes and to automate their training, you can hit a high success rate in record time, but you need an advanced platform with NLP included, a human in the loop, and all the tools and features you need to grow your AI assistant easily over time to do it. With an advanced conversational AI platform like ours, you don’t need to know anything about AI, LLMs, GPT-4, or LLaMA to instantly better your customer experience.
Watching people experiment with ChatGPT in particular, you might be nervous about accuracy. We are too. That’s why trusted and experienced AI providers won’t be suggesting LLM content feeds directly into your live customer conversations without your approval first. There’s potential for this in the future, but it’s a long way off.
We’ve put eight years’ worth of AI knowledge and 21+ years of data engineering experience into our AI Studio platform, so you can easily create and manage your own AI assistant while taking full advantage of advanced AI. Thanks to our human in the loop system and conversation review service, your AI assistant will operate at a safe level of accuracy from the minute it starts talking to your customers, since our expert team is always there to guide you.
We’ll continue to implement new LLM releases so you can safely benefit from them without the risk that comes with experimentation, and focus all your attention on giving your customers the most sophisticated level of support.