It has often been said that chat bots arrived too early—when technology wasn’t there yet. We all know that talking to them feels more like filling out a form than a conversation. They fail as soon as you go slightly off the script and are time-consuming to set up, requiring specialized libraries or tools.
Large text generation models like GPT-3 are a true breakthrough, but they cannot be served as a product by themselves for two main reasons. First, having been trained on vast quantities of diverse texts, they can drift away from their assigned topic and provide potentially misleading, incorrect, or even dangerous information. Second, a truly useful chat bot would not be based purely on general knowledge but would make use of a knowledge base specific to the particular task at hand.
That’s why we built our Quickchat bot engine on top of GPT-3. It takes care of:
- Ensuring the chat bot doesn’t drift away from the topic it was assigned
- Priming the chat bot with a knowledge base specific to a particular use case
How does it work?
Our chat bots are built based on unstructured knowledge (plain text, docs, FAQs, example conversations, etc.) for a specific use case. They can later be deployed on your website using our JavaScript widget or through a messaging service (e.g. Facebook Messenger, Instagram, Telegram, Slack) using our API.
Where do I start?
- Please get in touch at [email protected] and provide basic information about your use case (tasks the chat bot is expected to accomplish, description of the knowledge base to be provided, where it will be deployed, expected traffic).
- We will be in touch shortly to discuss further and arrange your Proof of Concept.