LivePerson will use ChatGPT AI in its chat platform
LivePerson, a global conversational platform company, is due to launch, by the end of March, an update that introduces artificial intelligence technologies from OpenAI, the creator of ChatGPT, to its conversational cloud, including the Generative Pre-Training Transformer 3. a system that essentially allows companies to create their own large language models (LLM) for building chatbots.
This is possible by feeding the platform with internal content, such as documents or websites. With this information, combined with general AI knowledge, the bot can be quickly configured to provide detailed customer service responses. This way, the company has more control over its chatbot, which is based on its own language model.
In a demonstration made during a LivePerson event, this Thursday, the 9th, Americo Talarico, the company’s Latin American sales consultant, showed that, simply by feeding the platform with information about a cruise, for example, the shipping company’s chatbot mariner can answer accurately. about activities aimed at children and all questions related to them, in natural language. Whether the caller asks the bot for holiday advice or not, it responds just as ChatGPT would.
“A lot of things revolve around data. Many of the conversations we had in Europe last week [no Mobile World Congress], were based on knowledge. How do we get this data into a large language model? This is really essential for us as an enterprise level company. We’re building a control interface that will allow brands to provide us with data from different knowledge bases, putting it into a big model language specifically for that brand,” said Adam Canter, LivePerson’s Americas SVP.
In order for the platform to comply with safety and reliability standards, avoiding the so-called “hallucinations” that bots like ChatGPT have experienced, giving answers considered absurd or wrong, it is necessary for people to review the questions and answers that pass conversations
Currently, on Conversational Cloud, there are 34 billion API calls per month and 1 billion conversational interactions per day. 300,000 agents enter it every day, from all over the world, who work in 18,000 companies, 80 of which are in Brazil. They can rate the quality of the content with a positive or negative sign, helping to train the chatbot, just like OpenAI itself does in its conversational interface. “Human feedback is going to be key for us when we start because we also start to scale. It’s going to happen quickly,” Canter said.
More AI features come with the update. One of them recommends actions the agent can take to resolve the service, based on data such as customer history. It is up to him to decide whether to follow them or not. It is as if the AI is an assistant, guiding the agent. Additionally, before passing the conversation to the person, the chatbot summarizes what has been discussed so far to better contextualize it in relation to the service.