Unit 4 - Chatbots and Digital Assistants

Generative AI (GenAI)

A type of artificial intelligence that can create new content, from text to images and music, by learning from vast amounts of existing data.

Natural Language Processing (NLP)

Branch of AI that helps computers understand, interpret, and manipulate human language. Interactions between computers and humans using natural language, in particular the programming of computers to effectively process large amounts of natural language data.

Supervised fine-tuning (SFT)

Supervised fine-tuning is the technique of further training a pre-trained machine learning model on a downstream task by updating the model parameters based on new labeled example data to improve performance on that specific task.

Alexa skills

Alexa's voice-driven capabilities. They can be enabled, disabled, and custom-developed.

Conversational AI

Set of technologies behind automated messaging and speech-enabled applications that offer human-like interactions between computers and humans, by recognizing speech and text, understanding intent, deciphering different languages, and responding in a way that mimics human conversation.

Smart Display

Smart speaker with touch screen attached to it.

Smart Speaker

Speaker and voice command device with an integrated virtual assistant (e.g. Amazon Alexa, Google Assistant, Apple's Siri) activated by a wake word and that offers interactive actions.

Wake word

Special word or phrase that is meant to activate a device when spoken, such as "OK Google", "Alexa", or "Hey Siri".

Self-prompting

The AI's ability to generate its own follow-up questions or prompts based on initial input, allowing for more dynamic and autonomous exploration or clarification of a topic.

Context Window

The maximum sequence length of contiguous text that a neural language model can ingest, reason over, and apply when generating textual outputs.

Turing Test

A test proposed in 1950 by Alan Turing - a famous English mathematician - in which a machine would be judged "intelligent" if the software could use conversation to fool a human into thinking it was talking with a person instead of a machine.

Multimodal AI

Multimodal AI refers to artificial intelligence systems that can understand, connect, and translate between multiple modes of data representation, like text, images, audio, video, and sensor inputs.

Prompt Engineering

The skillful design of input prompts for LLMs to produce high-quality, coherent outputs.

Chain-of-Thought prompting

A powerful technique that leverages a series of intermediate reasoning steps to enhance the abilities of large language models (LLMs) in tackling complex reasoning tasks. By breaking down the reasoning process into manageable stages, LLMs can generate more accurate answers and exhibit improved reasoning capabilities.

Chatbot

A program designed to carry on a conversation with a human user by mimicking written or spoken human speech to simulate a conversation or interaction with a real person.

Digital Assistant

A program using artificial intelligence and voice recognition software to understand voice commands.

Generative Pre-trained Transformer (GPT)

A type of deep learning model architecture used for natural language processing tasks. GPTs leverage self-supervised pretraining of transformer text encoders to gain impressive natural language generation capacities ranging from completion of text snippets to computer program generation and more based on previous context. Their generative nature and pretraining strategy enables producing high-quality, meaningful, on-topic text.

Parameters

Are the internal variables, such as weights, that the model uses to make predictions and which are adjusted during training based on the data.

Chatiquette

Etiquette dictating how one should behave in chat rooms.

Open Domain

Free to consider any kind of information you want. Free to talk about any topic or content.

Eavesdropping

Listening secretly to a conversation. The practice is widely regarded as unethical, and in many jurisdictions is illegal.

Blender

Name of Facebook's open-source and largest-ever open-domain AI chatbot at its time of introduction in April 2020.

Meena

Name of Google's AI chatbot which was introduced in February 2020.

Google Duplex

Name of the technology that was introduced in 2018, powering a Google Assistant feature, which could place calls on the behalf of a human with a natural-sounding human voice.

Eliza

Name of the very first chatbot using Natural Language Processing (NLP) created by MIT AI laboratory in 1965.

Messenger Chatbots

Pieces of software that uses AI to communicate with customers. They understand what is being asked and can then formulate a response in a very human way. Interesting fact: Facebook Messenger passed 300,000 bots in 2018.