It’s important not to confuse ChatGPT as being the complete term for all generative artificial intelligence technology, it is a chatbot built by OpenAI, a tech company. GPT and LLMs are broader definitions of the Generative AI technologies.
SentiOne algorithm storing the knowledge ensures the data is correct, valid and up-to-date, unlike the publicly available ChatGPT models. Generative AI is leveraged solely to generate an appropriate and tailored response from the multiple pieces of content. Using both of these technologies, we can assure accuracy in understanding the question (our models), and we can guarantee that the answers are generated from a validated and current data source.
Conversational AI: Intelligence that can communicate with humans in a natural way, where a computer can understand, process and generate natural language.
Generative AI: A type of AI that can create or generate text, images and video, depending on what a user may prompt or ask it to do.
GPT: Generative Pre-trained Transformers, a type of AI that uses deep learning, neural networks and large data sets to generate text. It is a type of LLM.
LLM: Large Language Model, is a language model containing many parameters trained on a large amount of text.
AI: Artificial Intelligence, computer systems that can do things humans can normally do.
NLU: Natural language understanding, a sub-field of natural language processing, which allows computers to understand natural language.
ChatGPT: An application built by OpenAI, a chatbot designed to converse with users to generate text-based answers. It has been tailored to answer conversationally, and is based on different models developed by OpenAI, from GPT-3 to GPT-4.
This diagram shows from a high level the components of the technologies involved. This shows a hybrid, or combination of Natural Language Processing and Generative AI technologies working together to process a user request through the Brexit Bot.
User input: this is the question or query coming in from the user as text.
Natural Language Processing
- A user’s query is assigned meaning and understood by the SentiOne NLU model through whats called intent classification.
- Based on the meaning of the query/topic, the NLU assigns the correct knowledge base to look for the information (a document, text, article or all content relating to the topic). This ensures we only use the correct and relevant content for our response.
Generative AI
- The type of character and response of the virtual assistant is outlined as a prompt for the Generative AI.
- The Generative AI model creates the reply to the user using all previous elements, or what is called Prompts. All together, this is called prompt chaining, where we feed the user’s query, the relevant knowledge base, the type of character it should respond as, and how the response should be styled.
Response: this is the generated response relevant to the user query in text.
