AI Glossary

Techniques & Methods

Adversarial Training

This approach enhances the resilience of machine learning models by exposing them to custom-crafted challenging examples. The idea is to make the model learn more effective and reliable ways to interpret data.

Applications

Agents

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This term refers to multi-purpose AI tools capable of performing various tasks, like Apple's Siri. More recently, agents based on Large Language Models (LLMs) have emerged, built on architectures like chain-of-thought. Notable examples include LangChain and AutoGPT."}" data-sheets-userformat="{"2":4354,"4":{"1":2,"2":16777215},"11":4,"15":"Arial"}">This term refers to multi-purpose AI tools capable of performing various tasks, like Apple's Siri. More recently, agents based on Large Language Models (LLMs) have emerged, built on architectures like chain-of-thought. Notable examples include LangChain and AutoGPT.

Core Concepts

AI (Artificial Intelligence)

AI is a subfield of computer science dedicated to creating machines that can perform activities normally requiring human intelligence, such as understanding speech, solving problems, and translating languages.

General

AI Trainer

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"These are individuals who assist in the refinement of AI models by offering feedback, ranking the quality of responses, and creating sample dialogues for training."}" data-sheets-userformat="{"2":4354,"4":{"1":2,"2":16777215},"11":4,"15":"Arial"}">These are individuals who assist in the refinement of AI models by offering feedback, ranking the quality of responses, and creating sample dialogues for training.

Core Concepts

Algorithm

Essentially, an algorithm is a set of rule-based instructions that a computer follows to perform a specific task or solve a problem.

Techniques & Methods

Alignment

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This refers to the process of fine-tuning an LLM to conform to certain ethical guidelines. It aims to mitigate issues arising from uncurated training data, such as biased or harmful outputs."}" data-sheets-userformat="{"2":4354,"4":{"1":2,"2":16777215},"11":4,"15":"Arial"}">This refers to the process of fine-tuning an LLM to conform to certain ethical guidelines. It aims to mitigate issues arising from uncurated training data, such as biased or harmful outputs.

Model Components

API (Application Programming Interface)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An API serves as a bridge allowing different software applications to communicate with each other. OpenAI, for example, offers an API that lets developers incorporate ChatGPT into their own platforms."}" data-sheets-userformat="{"2":4354,"4":{"1":2,"2":16777215},"11":4,"15":"Arial"}">An API serves as a bridge allowing different software applications to communicate with each other. OpenAI, for example, offers an API that lets developers incorporate ChatGPT into their own platforms.

Model Components

Artificial Neural Network

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"ANNs simulate human brain functionality by using interconnected processing units. Like neurons in a human brain, these units collectively work towards solving a problem or task, highlighting the cooperative aspect of both biological and artificial neural systems."}" data-sheets-userformat="{"2":4354,"4":{"1":2,"2":16777215},"11":4,"15":"Arial"}">ANNs simulate human brain functionality by using interconnected processing units. Like neurons in a human brain, these units collectively work towards solving a problem or task, highlighting the cooperative aspect of both biological and artificial neural systems.

Techniques & Methods

Attention

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"In the context of GPT's transformer architecture, attention refers to the model's ability to understand the relational context between words in a given text sequence. This technique was originally outlined in the paper \"Attention Is All You Need.\""}" data-sheets-userformat="{"2":4354,"4":{"1":2,"2":16777215},"11":4,"15":"Arial"}">In the context of GPT's transformer architecture, attention refers to the model's ability to understand the relational context between words in a given text sequence. This technique was originally outlined in the paper "Attention Is All You Need."

Techniques & Methods

Attention Mechanism

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This technique, employed in deep learning models, assigns varying degrees of significance or \"attention\" to different words or features during data processing."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This technique, employed in deep learning models, assigns varying degrees of significance or "attention" to different words or features during data processing.

Core Concepts

Augmented Intelligence

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Augmented Intelligence combines human intuition and machine computation power to make more effective decisions. Unlike approaches that seek to replace humans with machines, augmented intelligence aims to provide a synergistic interaction between the two."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Augmented Intelligence combines human intuition and machine computation power to make more effective decisions. Unlike approaches that seek to replace humans with machines, augmented intelligence aims to provide a synergistic interaction between the two.

Core Concepts

Autonomous

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"If a machine can operate independently without human guidance or intervention, it is considered autonomous."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">If a machine can operate independently without human guidance or intervention, it is considered autonomous.

Techniques & Methods

Autoregression

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"In language models, this refers to the dependency of future token predictions on the tokens previously generated by the model, not just the initial input."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">In language models, this refers to the dependency of future token predictions on the tokens previously generated by the model, not just the initial input.

Model Components

Autoregressive Model

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This is a statistical model that forecasts future values based on past, time-lagged data points. In the context of ChatGPT, it predicts the subsequent word in a sentence using this approach."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This is a statistical model that forecasts future values based on past, time-lagged data points. In the context of ChatGPT, it predicts the subsequent word in a sentence using this approach.

Techniques & Methods

Backpropagation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This algorithm is used to train neural networks by optimizing the weights of the network in relation to the gradient of a loss function. It is essential for accurately fine-tuning the model."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This algorithm is used to train neural networks by optimizing the weights of the network in relation to the gradient of a loss function. It is essential for accurately fine-tuning the model.

Techniques & Methods

Backward Chaining

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A reasoning approach where the algorithm starts from a desired outcome and works its way backward to identify the data or conditions that could lead to that outcome."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A reasoning approach where the algorithm starts from a desired outcome and works its way backward to identify the data or conditions that could lead to that outcome.

Techniques & Methods

Bandit Optimization

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A real-time decision-making strategy in machine learning that balances the act of exploring new options with exploiting known effective options, all while having limited information."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A real-time decision-making strategy in machine learning that balances the act of exploring new options with exploiting known effective options, all while having limited information.

Techniques & Methods

Beam Search

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"In natural language processing (NLP), this is an algorithm that sifts through sequences of tokens to find the most probable sequence. It does so by keeping track of a fixed number of the best sequences at each step."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">In natural language processing (NLP), this is an algorithm that sifts through sequences of tokens to find the most probable sequence. It does so by keeping track of a fixed number of the best sequences at each step.

Core Concepts

Bias

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"In machine learning, this refers to the assumptions a model makes to simplify the learning process. Lower bias is often desirable in supervised learning, as these assumptions can hinder performance."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">In machine learning, this refers to the assumptions a model makes to simplify the learning process. Lower bias is often desirable in supervised learning, as these assumptions can hinder performance.

General

Big Data

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Refers to datasets that are too voluminous or complicated to be handled by conventional data-processing applications."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Refers to datasets that are too voluminous or complicated to be handled by conventional data-processing applications.

Model Components

Bounding Box

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"In the context of image or video analysis, a bounding box is an imaginary rectangle used to highlight specific areas and label them, aiding the model in object recognition."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">In the context of image or video analysis, a bounding box is an imaginary rectangle used to highlight specific areas and label them, aiding the model in object recognition.

Techniques & Methods

Chain-of-Thought

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A prompting technique designed to enhance the reasoning abilities of Large Language Models (LLMs). The technique breaks down complex tasks or questions into simpler steps for better understanding, and may even involve additional data gathering through web searches. A more advanced version, known as Tree-of-Thought, allows for logical branching and backtracking."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A prompting technique designed to enhance the reasoning abilities of Large Language Models (LLMs). The technique breaks down complex tasks or questions into simpler steps for better understanding, and may even involve additional data gathering through web searches. A more advanced version, known as Tree-of-Thought, allows for logical branching and backtracking.

Applications

Chatbot

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Software designed to engage in text or voice conversations with humans, emulating human-like interaction."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Software designed to engage in text or voice conversations with humans, emulating human-like interaction.

Applications

ChatGPT

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Introduced in 2022, ChatGPT is designed to maintain a conversational state, making it capable of addressing follow-up questions within a dialogue, unlike its stateless GPT predecessors."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Introduced in 2022, ChatGPT is designed to maintain a conversational state, making it capable of addressing follow-up questions within a dialogue, unlike its stateless GPT predecessors.

Core Concepts

Cognitive Computing

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Often used interchangeably with artificial intelligence, this term is preferred by some to sidestep the sci-fi connotations that AI may have."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Often used interchangeably with artificial intelligence, this term is preferred by some to sidestep the sci-fi connotations that AI may have.

Techniques & Methods

Completion

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The generated output from a Large Language Model (LLM) in response to a given prompt. This term originates from the way prompts are usually structured, requiring the model to complete a sentence or thought."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The generated output from a Large Language Model (LLM) in response to a given prompt. This term originates from the way prompts are usually structured, requiring the model to complete a sentence or thought.

Core Concepts

Computational Learning Theory

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A subfield within AI focused on the formulation and analysis of algorithms used in machine learning."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A subfield within AI focused on the formulation and analysis of algorithms used in machine learning.

Model Components

Contextual Embeddings

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"These are word embeddings created by considering surrounding words or sentence structure, making them sensitive to the context in which a word appears."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">These are word embeddings created by considering surrounding words or sentence structure, making them sensitive to the context in which a word appears.

Model Components

Context Window

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"For ChatGPT, this refers to the extent of recent dialogue history that the model can use to generate appropriate responses."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">For ChatGPT, this refers to the extent of recent dialogue history that the model can use to generate appropriate responses.

Techniques & Methods

Coreference Resolution

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The task of identifying and harmonizing different expressions in a text that refer to the same entity."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The task of identifying and harmonizing different expressions in a text that refer to the same entity.

Miscellaneous

Corpus

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A comprehensive collection of text or speech data, typically used to train machines for linguistic tasks."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A comprehensive collection of text or speech data, typically used to train machines for linguistic tasks.

Applications

CRM with AI

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Customer Relationship Management (CRM) technology integrated with AI capabilities can provide advanced functionalities like pre-composed personalized emails, automated customer service replies, and enhanced marketing strategies."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Customer Relationship Management (CRM) technology integrated with AI capabilities can provide advanced functionalities like pre-composed personalized emails, automated customer service replies, and enhanced marketing strategies.

Techniques & Methods

Data Augmentation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The practice of increasing the volume of training data by adding variations or creating synthetic data based on existing examples."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The practice of increasing the volume of training data by adding variations or creating synthetic data based on existing examples.

Techniques & Methods

Data Mining

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The activity of examining datasets to unearth novel patterns, which may enhance model performance."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The activity of examining datasets to unearth novel patterns, which may enhance model performance.

Miscellaneous

Data Privacy

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"In the context of ChatGPT, this pertains to the guarantee that conversations are kept confidential and not stored beyond a 30-day period."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">In the context of ChatGPT, this pertains to the guarantee that conversations are kept confidential and not stored beyond a 30-day period.

Miscellaneous

Data Science

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An interdisciplinary field leveraging statistical, computational, and informational techniques to solve data-centric problems."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">An interdisciplinary field leveraging statistical, computational, and informational techniques to solve data-centric problems.

Miscellaneous

Dataset

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An organized collection of related data points, usually categorized with labels and specific structures."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">An organized collection of related data points, usually categorized with labels and specific structures.

Techniques & Methods

Decoding Rules

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Guidelines that govern the process of text generation in a language model."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Guidelines that govern the process of text generation in a language model.

Core Concepts

Deep Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A sophisticated AI technique that enables computers to recognize intricate patterns in data by using multi-layered neural networks. This is particularly useful in tasks such as image and speech recognition, as well as natural language understanding."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A sophisticated AI technique that enables computers to recognize intricate patterns in data by using multi-layered neural networks. This is particularly useful in tasks such as image and speech recognition, as well as natural language understanding.

Techniques & Methods

Dependency Parsing

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"In natural language processing, this is the act of dissecting a sentence to discern the grammatical connections between its various elements, like subjects, objects, and modifiers."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">In natural language processing, this is the act of dissecting a sentence to discern the grammatical connections between its various elements, like subjects, objects, and modifiers.

Miscellaneous

Deployment

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This involves making a fully trained language model operational, either by incorporating it into an existing system or offering an API for external access."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This involves making a fully trained language model operational, either by incorporating it into an existing system or offering an API for external access.

Applications

Dialogue System

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A computational framework designed for simulating human-like conversation."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A computational framework designed for simulating human-like conversation.

Model Components

Discriminator (in GAN)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"In a Generative Adversarial Network, the discriminator serves as a sort of \"quality control.\" It distinguishes between genuine data from a dataset and fabricated data produced by the generator, its counter-part in the network."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">In a Generative Adversarial Network, the discriminator serves as a sort of "quality control." It distinguishes between genuine data from a dataset and fabricated data produced by the generator, its counter-part in the network.

Techniques & Methods

Distributed Training

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The technique of spreading the training of an AI model across multiple computing devices, enabling quicker processing and greater data capacity."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The technique of spreading the training of an AI model across multiple computing devices, enabling quicker processing and greater data capacity.

Model Components

Embeddings

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Unlike simple tokens, embeddings numerically represent text while capturing its semantic essence. This means that words with similar meanings will have similar embeddings."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Unlike simple tokens, embeddings numerically represent text while capturing its semantic essence. This means that words with similar meanings will have similar embeddings.

Model Components

Encoder

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This is a segment of a neural network specialized in processing incoming data to identify significant features. In the case of ChatGPT, encoders help the model understand the prompt and generate a fitting response."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This is a segment of a neural network specialized in processing incoming data to identify significant features. In the case of ChatGPT, encoders help the model understand the prompt and generate a fitting response.

Applications

Enterprise AI

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The application of AI for delivering predictive analytics to professionals, aiming to optimize organizational processes and operations."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The application of AI for delivering predictive analytics to professionals, aiming to optimize organizational processes and operations.

Core Concepts

Entities

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"These are identifiable real-world elements like individuals, groups, places, or things that can be recognized and extracted from textual data."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">These are identifiable real-world elements like individuals, groups, places, or things that can be recognized and extracted from textual data.

Techniques & Methods

Entity Annotation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This involves tagging unstructured text with labels that identify specific entities like people, organizations, or locations, making the text machine-readable."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This involves tagging unstructured text with labels that identify specific entities like people, organizations, or locations, making the text machine-readable.

Techniques & Methods

Entity Extraction

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A broad term describing the techniques used to add structure to data, usually via annotation, so it becomes interpretable by machine learning algorithms."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A broad term describing the techniques used to add structure to data, usually via annotation, so it becomes interpretable by machine learning algorithms.

Miscellaneous

Ethical AI Maturity Model

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A framework for organizations to assess and evolve their ethical standing in the deployment of AI technologies. It deals with multiple dimensions including transparency, fairness, data privacy, accountability, and predictive bias."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A framework for organizations to assess and evolve their ethical standing in the deployment of AI technologies. It deals with multiple dimensions including transparency, fairness, data privacy, accountability, and predictive bias.

Techniques & Methods

Evaluation Metrics

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Quantifiable standards like perplexity, accuracy, F1 score, or BLEU score, used to gauge the effectiveness of a language model."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Quantifiable standards like perplexity, accuracy, F1 score, or BLEU score, used to gauge the effectiveness of a language model.

Core Concepts

Explainable AI (XAI)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This refers to AI models designed to make their decision-making processes understandable to human users."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This refers to AI models designed to make their decision-making processes understandable to human users.

Techniques & Methods

Extractive Summarization

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This entails generating a summary by selecting key sentences or points from the original text. ChatGPT can perform this function when summarizing large volumes of text."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This entails generating a summary by selecting key sentences or points from the original text. ChatGPT can perform this function when summarizing large volumes of text.

Techniques & Methods

Feature Extraction

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This involves isolating essential pieces of information from a dataset for easier processing and analysis, a step ChatGPT utilizes to better understand prompts and formulate responses."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This involves isolating essential pieces of information from a dataset for easier processing and analysis, a step ChatGPT utilizes to better understand prompts and formulate responses.

Techniques & Methods

Few-Shot Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The ability of a model to understand a task quickly with very few training examples."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The ability of a model to understand a task quickly with very few training examples.

Techniques & Methods

Fine-Grained Control

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This refers to the capability of a model to produce text with specific characteristics, such as a particular tone or style, by altering its internal settings."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This refers to the capability of a model to produce text with specific characteristics, such as a particular tone or style, by altering its internal settings.

Techniques & Methods

Fine Tuning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The process of adapting a pre-trained model to more specific tasks by training it further on a specialized dataset. This is generally more efficient than starting training from scratch."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The process of adapting a pre-trained model to more specific tasks by training it further on a specialized dataset. This is generally more efficient than starting training from scratch.

Techniques & Methods

Forward Chaining

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This is a reasoning technique where the AI starts with available data to form possible solutions, assessing various hypotheses to identify those most relevant to the issue at hand."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This is a reasoning technique where the AI starts with available data to form possible solutions, assessing various hypotheses to identify those most relevant to the issue at hand.

Model Components

Foundational Model

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"These are large-scale models trained on diverse datasets, enabling them to be applicable to a wide array of tasks. The creation of such models is often both expensive and time-consuming."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">These are large-scale models trained on diverse datasets, enabling them to be applicable to a wide array of tasks. The creation of such models is often both expensive and time-consuming.

Core Concepts

General AI

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Also known as strong AI, this term describes an AI capable of performing any cognitive task that a human can do."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Also known as strong AI, this term describes an AI capable of performing any cognitive task that a human can do.

Techniques & Methods

Generation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Refers to the act of creating new text using a language model. This can be automated via the model's own token predictions or guided by human input."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Refers to the act of creating new text using a language model. This can be automated via the model's own token predictions or guided by human input.

Model Components

Generative Adversarial Network (GAN)

A GAN consists of two interlocking neural networks.

Core Concepts

Generative AI

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This AI subfield concentrates on creating novel content derived from existing data. In a Customer Relationship Management (CRM) setting, generative AI can produce anything from personalized marketing copy to simulated data for strategy testing."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This AI subfield concentrates on creating novel content derived from existing data. In a Customer Relationship Management (CRM) setting, generative AI can produce anything from personalized marketing copy to simulated data for strategy testing.

Model Components

Generative Model

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A machine learning model engineered to create new data resembling the data it was trained on. ChatGPT, for instance, leverages generative models to produce coherent and contextually relevant responses."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A machine learning model engineered to create new data resembling the data it was trained on. ChatGPT, for instance, leverages generative models to produce coherent and contextually relevant responses.

Model Components

Generative Pre-trained Transformer (GPT)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Part of a neural network family designed for content creation, GPT models are pre-trained on extensive text datasets, allowing them to generate coherent and contextually relevant text based on queries or prompts."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Part of a neural network family designed for content creation, GPT models are pre-trained on extensive text datasets, allowing them to generate coherent and contextually relevant text based on queries or prompts.

Model Components

Generator

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"In AI, a generator is a tool that fabricates new content in response to user input. It learns from existing data to produce output that follows the same patterns and features. ChatGPT is a renowned example of a text generator."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">In AI, a generator is a tool that fabricates new content in response to user input. It learns from existing data to produce output that follows the same patterns and features. ChatGPT is a renowned example of a text generator.

Model Components

GPT-3 (Generative Pre-trained Transformer 3)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A specific AI language model from OpenAI, GPT-3 is trained on a vast corpus of web text to generate text that closely resembles human-written content."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A specific AI language model from OpenAI, GPT-3 is trained on a vast corpus of web text to generate text that closely resembles human-written content.

Techniques & Methods

Greedy Algorithms

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"These algorithms, often used in NLP, aim to find the most probable sequence of tokens by selecting the token with the highest likelihood at each step."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">These algorithms, often used in NLP, aim to find the most probable sequence of tokens by selecting the token with the highest likelihood at each step.

Techniques & Methods

Hallucination

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This occurs when a generative AI model produces content that deviates from factual or logical expectations. For example, generating an animal image that combines disparate features from different species. Such outcomes are usually considered errors."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This occurs when a generative AI model produces content that deviates from factual or logical expectations. For example, generating an animal image that combines disparate features from different species. Such outcomes are usually considered errors.

Techniques & Methods

Heuristics

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"These are general guidelines or tactics employed for decision-making or problem-solving. ChatGPT uses heuristics to understand input and generate appropriate output."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">These are general guidelines or tactics employed for decision-making or problem-solving. ChatGPT uses heuristics to understand input and generate appropriate output.

Core Concepts

Hyperparameter

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Though sometimes used interchangeably with \"parameter,\" hyperparameters are set manually and influence how the model learns from the data."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Though sometimes used interchangeably with "parameter," hyperparameters are set manually and influence how the model learns from the data.

Techniques & Methods

Inference

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The act of utilizing a trained model to make predictions or decisions based on new, unseen data."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The act of utilizing a trained model to make predictions or decisions based on new, unseen data.

Techniques & Methods

Information Extraction

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This is the technique of pulling out structured information from unstructured sources like text or images. It is a process ChatGPT employs to understand user input and formulate relevant responses."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This is the technique of pulling out structured information from unstructured sources like text or images. It is a process ChatGPT employs to understand user input and formulate relevant responses.

Applications

InstructGPT

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An enhanced version of ChatGPT, specifically designed to follow task instructions in a prompt and provide comprehensive explanations."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">An enhanced version of ChatGPT, specifically designed to follow task instructions in a prompt and provide comprehensive explanations.

Core Concepts

Intent

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"In the context of NLP and chatbots, intent refers to the underlying purpose of a given phrase or sentence, such as the desire to reduce volume when someone says, \"turn the volume down.\""}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">In the context of NLP and chatbots, intent refers to the underlying purpose of a given phrase or sentence, such as the desire to reduce volume when someone says, "turn the volume down."

Techniques & Methods

Joint Probability

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This is the likelihood of multiple events happening simultaneously. ChatGPT uses joint probability calculations to understand and respond to user input."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This is the likelihood of multiple events happening simultaneously. ChatGPT uses joint probability calculations to understand and respond to user input.

Miscellaneous

Knowledge Base

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This is a structured collection of information, often used to supplement and contextualize the data that language models work with."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This is a structured collection of information, often used to supplement and contextualize the data that language models work with.

Techniques & Methods

Knowledge Representation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The practice of formally structuring knowledge so that it is machine-readable. ChatGPT employs this to better understand and respond to queries."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The practice of formally structuring knowledge so that it is machine-readable. ChatGPT employs this to better understand and respond to queries.

Miscellaneous

Label

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A tag in training data that denotes the expected output for a specific data point."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A tag in training data that denotes the expected output for a specific data point.

Model Components

Language Model

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A model built on mathematical and statistical principles to predict subsequent words or word sequences in a text."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A model built on mathematical and statistical principles to predict subsequent words or word sequences in a text.

Model Components

Large Language Model (LLM)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A type of AI trained on extensive textual data, capable of generating human-like text based on provided prompts. LLMs have a wide range of applications, from answering queries to creative writing."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A type of AI trained on extensive textual data, capable of generating human-like text based on provided prompts. LLMs have a wide range of applications, from answering queries to creative writing.

Core Concepts

Latent Variables

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"These are unobserved variables inferred from observed data. ChatGPT utilizes latent variable models to better understand and respond to user input."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">These are unobserved variables inferred from observed data. ChatGPT utilizes latent variable models to better understand and respond to user input.

Techniques & Methods

Linguistic Annotation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The act of tagging sentence data sets with particular subjects, often for purposes like sentiment analysis or other NLP tasks."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The act of tagging sentence data sets with particular subjects, often for purposes like sentiment analysis or other NLP tasks.

Techniques & Methods

Low Rank Adaption (LoRA)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A fine-tuning technique that is both memory-efficient and cost-effective, allowing for quicker model training."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A fine-tuning technique that is both memory-efficient and cost-effective, allowing for quicker model training.

Core Concepts

Machine Intelligence

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A collective term that includes various learning algorithms like machine learning and deep learning."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A collective term that includes various learning algorithms like machine learning and deep learning.

Core Concepts

Machine Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A subfield of AI where computers learn from data without being explicitly programmed. Just like a child learns to identify animals from pictures and feedback, machine learning algorithms use labeled data to make accurate decisions and predictions."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A subfield of AI where computers learn from data without being explicitly programmed. Just like a child learns to identify animals from pictures and feedback, machine learning algorithms use labeled data to make accurate decisions and predictions.

Applications

Machine Translation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Automated conversion of text from one language to another without human intervention."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Automated conversion of text from one language to another without human intervention.

Techniques & Methods

Markov Decision Process

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A mathematical structure for modeling decision-making scenarios where the result of each decision relies on the current state. This framework is employed by ChatGPT for interpreting inputs and producing fitting responses."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A mathematical structure for modeling decision-making scenarios where the result of each decision relies on the current state. This framework is employed by ChatGPT for interpreting inputs and producing fitting responses.

Techniques & Methods

Masked Language Modeling

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A training technique where selected tokens in an input sequence are obscured, requiring the model to predict them based on the remaining contextual information."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A training technique where selected tokens in an input sequence are obscured, requiring the model to predict them based on the remaining contextual information.

Model Components

Maximum Response Length

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The upper bound on the amount of text ChatGPT can produce in a single reply."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The upper bound on the amount of text ChatGPT can produce in a single reply.

Model Components

Model

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A computational entity trained to identify data patterns, which can then perform tasks like weather prediction, image recognition, and language translation."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A computational entity trained to identify data patterns, which can then perform tasks like weather prediction, image recognition, and language translation.

Model Components

Model Architecture

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The specific arrangement and functionality of individual neural network components within a complex AI model. Examples include convolutional networks, transformers, and recurrent networks."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The specific arrangement and functionality of individual neural network components within a complex AI model. Examples include convolutional networks, transformers, and recurrent networks.

Miscellaneous

Model Card

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A document offering in-depth insights into a machine learning model, such as its capabilities, limitations, and ideal applications."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A document offering in-depth insights into a machine learning model, such as its capabilities, limitations, and ideal applications.

Applications

Moderation Tools

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Utilities made available to developers for governing the behavior of AI models within applications or services."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Utilities made available to developers for governing the behavior of AI models within applications or services.

Core Concepts

Multi-modal AI

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"AI systems capable of understanding and processing different forms of inputs, like text, images, speech, and videos."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">AI systems capable of understanding and processing different forms of inputs, like text, images, speech, and videos.

Techniques & Methods

Multitask Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The technique of training a single model on various tasks at once, enhancing its performance and shared learning across different tasks."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The technique of training a single model on various tasks at once, enhancing its performance and shared learning across different tasks.

Applications

Multi-turn Dialogue

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Interactive conversations that involve an ongoing exchange between two parties, such as between a user and an AI system."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Interactive conversations that involve an ongoing exchange between two parties, such as between a user and an AI system.

Techniques & Methods

Named Entity Recognition (NER)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An NLP task that involves identifying and categorizing named entities in text, like names of people, places, or organizations."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">An NLP task that involves identifying and categorizing named entities in text, like names of people, places, or organizations.

Core Concepts

Natural Language Generation (NLG)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The computational process of converting structured data into text or speech that is easily understandable by humans."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The computational process of converting structured data into text or speech that is easily understandable by humans.

Core Concepts

Natural Language Processing (NLP)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An AI field that specializes in enabling computers to understand, interpret, and produce human languages."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">An AI field that specializes in enabling computers to understand, interpret, and produce human languages.

Core Concepts

Natural Language Understanding (NLU)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A subfield of NLP focused on assisting machines in grasping the implicit meanings in human language, including context and errors."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A subfield of NLP focused on assisting machines in grasping the implicit meanings in human language, including context and errors.

Model Components

Neural Network

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Also known as Artificial Neural Networks (ANN), these are AI algorithms modeled after the human brain, consisting of numerous interconnected nodes."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Also known as Artificial Neural Networks (ANN), these are AI algorithms modeled after the human brain, consisting of numerous interconnected nodes.

Techniques & Methods

Offline Reinforcement Learning (RL)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A technique for training AI models using a pre-collected dataset, without needing real-time environmental interactions."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A technique for training AI models using a pre-collected dataset, without needing real-time environmental interactions.

Techniques & Methods

One-Shot / Few-Shot

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Terminology for prompts that either ask a straightforward question (one-shot) or require multiple examples for improved performance (few-shot)."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Terminology for prompts that either ask a straightforward question (one-shot) or require multiple examples for improved performance (few-shot).

Techniques & Methods

One-Shot Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The capability of a model to understand a new task from just a single example during its training phase."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The capability of a model to understand a new task from just a single example during its training phase.

Techniques & Methods

Online Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A machine learning approach where the model gets trained incrementally as new data becomes available."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A machine learning approach where the model gets trained incrementally as new data becomes available.

Miscellaneous

OpenAI

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The research organization behind the development of GPT-3 and ChatGPT, focused on ensuring artificial general intelligence benefits humanity broadly."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The research organization behind the development of GPT-3 and ChatGPT, focused on ensuring artificial general intelligence benefits humanity broadly.

Core Concepts

Overfitting

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A machine learning issue where a model performs exceptionally well on training data but poorly on new, unseen data."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A machine learning issue where a model performs exceptionally well on training data but poorly on new, unseen data.

Techniques & Methods

Overuse Penalty

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A corrective measure applied in ChatGPT’s decoding algorithm to curb repetitive phrasing."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A corrective measure applied in ChatGPT’s decoding algorithm to curb repetitive phrasing.

Model Components

Parameter

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An internal variable in the model used for making predictions, which is determined through the training data."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">An internal variable in the model used for making predictions, which is determined through the training data.

Techniques & Methods

Part-of-Speech Tagging (POS)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An NLP task where each word in a sentence is labeled according to its grammatical function, such as a noun or verb."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">An NLP task where each word in a sentence is labeled according to its grammatical function, such as a noun or verb.

Core Concepts

Pattern Recognition

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The field concerned with detecting regularities and structures within data."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The field concerned with detecting regularities and structures within data.

Applications

Plugins / Tools

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Additional capabilities that large language models can utilize through API access, enhancing their functional range and reducing the likelihood of errors."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Additional capabilities that large language models can utilize through API access, enhancing their functional range and reducing the likelihood of errors.

Applications

Predictive Analytics

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A discipline that merges data mining and machine learning to forecast future events based on historical data."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A discipline that merges data mining and machine learning to forecast future events based on historical data.

Model Components

Predictive Model

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A machine learning model designed to forecast the probability of specific outcomes based on provided input data. ChatGPT, for example, employs predictive models to tailor its replies to user input."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A machine learning model designed to forecast the probability of specific outcomes based on provided input data. ChatGPT, for example, employs predictive models to tailor its replies to user input.

Techniques & Methods

Pre-training in AI

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The initial phase of training a language model on a massive dataset before honing it for specialized tasks."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The initial phase of training a language model on a massive dataset before honing it for specialized tasks.

Techniques & Methods

Prompt

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The textual stimulus presented to a large language model (LLM). Prompts can range from short queries, like \"What's the capital of England?\", to lengthy textual descriptions of problems or tasks. Models often have token limitations, affecting the length of acceptable prompts."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The textual stimulus presented to a large language model (LLM). Prompts can range from short queries, like "What's the capital of England?", to lengthy textual descriptions of problems or tasks. Models often have token limitations, affecting the length of acceptable prompts.

Techniques & Methods

Prompt Engineering

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The art of carefully crafting or selecting the query or statement fed into a machine learning model in order to yield a specific, desired output."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The art of carefully crafting or selecting the query or statement fed into a machine learning model in order to yield a specific, desired output.

Techniques & Methods

Prompt Injection

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A technique for influencing or manipulating the behavior of products like chatbots or GitHub Copilot that employ LLM technology. This can reveal internal system settings or undermine alignment training."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A technique for influencing or manipulating the behavior of products like chatbots or GitHub Copilot that employ LLM technology. This can reveal internal system settings or undermine alignment training.

Techniques & Methods

Proximal Policy Optimization (PPO)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An algorithm employed in reinforcement learning to streamline the training process for models."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">An algorithm employed in reinforcement learning to streamline the training process for models.

Miscellaneous

Python

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A widely-used programming language versatile enough for various computing tasks."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A widely-used programming language versatile enough for various computing tasks.

Applications

QA (Question Answering)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An NLP task where a model responds to a posed question with a coherent and relevant answer based on its linguistic understanding and stored knowledge."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">An NLP task where a model responds to a posed question with a coherent and relevant answer based on its linguistic understanding and stored knowledge.

Techniques & Methods

Query

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A request made to a database or other source of information to extract data. ChatGPT utilizes queries to pull pertinent information for generating responses."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A request made to a database or other source of information to extract data. ChatGPT utilizes queries to pull pertinent information for generating responses.

Model Components

Recurrent Neural Network (RNN)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A neural network specialized in handling sequential data types like text. ChatGPT employs RNNs to interpret incoming user input and generate fitting replies."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A neural network specialized in handling sequential data types like text. ChatGPT employs RNNs to interpret incoming user input and generate fitting replies.

Techniques & Methods

Regularization

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A method for curtailing overfitting by incorporating a penalty term into the loss function to limit model complexity."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A method for curtailing overfitting by incorporating a penalty term into the loss function to limit model complexity.

Core Concepts

Reinforcement Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A type of machine learning that sets generalized objectives, encouraging the model to explore and experiment with different approaches rather than strictly adhering to pre-defined metrics."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A type of machine learning that sets generalized objectives, encouraging the model to explore and experiment with different approaches rather than strictly adhering to pre-defined metrics.

Techniques & Methods

Reinforcement Learning from Human Feedback (RLHF)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A technique for aligning models by using human feedback to train a reward model, subsequently fine-tuning a large language model."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A technique for aligning models by using human feedback to train a reward model, subsequently fine-tuning a large language model.

Techniques & Methods

Response Quality

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An evaluation metric considering how well an AI system replies to prompts, taking into account the relevance, coherence, and accuracy of the response."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">An evaluation metric considering how well an AI system replies to prompts, taking into account the relevance, coherence, and accuracy of the response.

Techniques & Methods

Retrieval Augmented Generation (RAG)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The practice of enriching a model's prompt with additional information obtained from web searches or private documents to generate a more informed output."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The practice of enriching a model's prompt with additional information obtained from web searches or private documents to generate a more informed output.

Model Components

Retrieval Model

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A system, often a Transformer, that fetches information from a data source. Pairing retrieval models with large language models can anchor the generated output in a factual context."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A system, often a Transformer, that fetches information from a data source. Pairing retrieval models with large language models can anchor the generated output in a factual context.

Model Components

Reward Models

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Models specifically designed to evaluate and rank various possible responses generated by a language model."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Models specifically designed to evaluate and rank various possible responses generated by a language model.

Miscellaneous

Sandbox Environment

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A controlled digital space where developers can experiment with and test new code without impacting the live application."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A controlled digital space where developers can experiment with and test new code without impacting the live application.

Techniques & Methods

Scaling Laws

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"An observed principle in AI stating that model performance typically improves with increases in data, computational power, and model size."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">An observed principle in AI stating that model performance typically improves with increases in data, computational power, and model size.

Techniques & Methods

Self-Attention

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A mechanism within Transformer-based models allowing them to discern relationships among tokens in an input sequence, thereby informing the output generation process."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A mechanism within Transformer-based models allowing them to discern relationships among tokens in an input sequence, thereby informing the output generation process.

Techniques & Methods

Semantic Annotation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The act of tagging queries or products with relevant markers to improve search engine accuracy."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The act of tagging queries or products with relevant markers to improve search engine accuracy.

Applications

Semantic Search

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A search approach that focuses on understanding user intent and contextual meanings of search terms to enhance result relevance."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A search approach that focuses on understanding user intent and contextual meanings of search terms to enhance result relevance.

Techniques & Methods

Semantic Similarity

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A metric evaluating the degree of meaning similarity between two text snippets, usually in a vector space."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A metric evaluating the degree of meaning similarity between two text snippets, usually in a vector space.

Applications

Sentiment Analysis

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A method to gauge the emotional nuance in textual content, often used to analyze customer feedback or social media discourse surrounding a brand or product."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A method to gauge the emotional nuance in textual content, often used to analyze customer feedback or social media discourse surrounding a brand or product.

Techniques & Methods

Sequence Generation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The act of using a language model to produce a chain of tokens, which could form a sentence or paragraph, based on learned patterns."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The act of using a language model to produce a chain of tokens, which could form a sentence or paragraph, based on learned patterns.

Model Components

Sequence-to-Sequence (Seq2Seq) Models

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Neural network designs specialized in translating an input sequence to a corresponding output sequence, common in applications like machine translation and text summarization."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Neural network designs specialized in translating an input sequence to a corresponding output sequence, common in applications like machine translation and text summarization.

Core Concepts

Strong AI

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Research aimed at developing AI with cognitive abilities comparable to human intelligence, often synonymous with General AI."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Research aimed at developing AI with cognitive abilities comparable to human intelligence, often synonymous with General AI.

Techniques & Methods

Supervised Fine-Tuning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The preliminary stage of fine-tuning in which human AI trainers supply dialogues involving both the user and the AI to the model."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The preliminary stage of fine-tuning in which human AI trainers supply dialogues involving both the user and the AI to the model.

Core Concepts

Supervised Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A learning paradigm where a model is trained on labeled examples, akin to a student learning from a teacher. This method is commonly used in tasks such as image recognition, language translation, and outcome prediction."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A learning paradigm where a model is trained on labeled examples, akin to a student learning from a teacher. This method is commonly used in tasks such as image recognition, language translation, and outcome prediction.

Miscellaneous

System Message

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The introductory text displayed when users initiate a conversation with ChatGPT."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The introductory text displayed when users initiate a conversation with ChatGPT.

Techniques & Methods

System Prompt

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A built-in prompt that accompanies user input to set the behavior of a chatbot or other LLM applications. Instances of system prompt leaks have offered insights into how these tools are configured."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A built-in prompt that accompanies user input to set the behavior of a chatbot or other LLM applications. Instances of system prompt leaks have offered insights into how these tools are configured.

Miscellaneous

Test Data

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Unlabeled data used to evaluate a machine learning model's ability to execute its designated task."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Unlabeled data used to evaluate a machine learning model's ability to execute its designated task.

Techniques & Methods

Text Classification

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The categorization of textual content based on its semantic content. ChatGPT utilizes text classification to understand user queries and generate suitable responses."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The categorization of textual content based on its semantic content. ChatGPT utilizes text classification to understand user queries and generate suitable responses.

Core Concepts

Token

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A numerical representation of text, enabling it to be processed by neural networks. Tokens can vary in size, from individual letters to whole words."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A numerical representation of text, enabling it to be processed by neural networks. Tokens can vary in size, from individual letters to whole words.

Techniques & Methods

Topic Modeling

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A method for identifying prevalent themes across a set of textual documents by analyzing word co-occurrence patterns."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A method for identifying prevalent themes across a set of textual documents by analyzing word co-occurrence patterns.

Techniques & Methods

Training

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The process of exposing a neural network to vast datasets to adjust its parameters and improve its output. Large model training involves enormous datasets and can take significant time."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The process of exposing a neural network to vast datasets to adjust its parameters and improve its output. Large model training involves enormous datasets and can take significant time.

Miscellaneous

Training Data

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The dataset used specifically for training a machine learning algorithm, as opposed to testing or validation."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The dataset used specifically for training a machine learning algorithm, as opposed to testing or validation.

Techniques & Methods

Transfer Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A technique involving the training of a machine on a related task to enhance its performance on its primary task."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A technique involving the training of a machine on a related task to enhance its performance on its primary task.

Model Components

Transformer

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A deep learning model type particularly adept at handling language data. Transformers excel in contextual understanding and are named for their ability to transform input data into meaningful output."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A deep learning model type particularly adept at handling language data. Transformers excel in contextual understanding and are named for their ability to transform input data into meaningful output.

Model Components

Transformer Decoder

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A component of the transformer model responsible for predicting subsequent tokens in a given sequence."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A component of the transformer model responsible for predicting subsequent tokens in a given sequence.

Model Components

Transformers

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A class of neural network architectures used in NLP, specifically designed to process sequences, including text."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A class of neural network architectures used in NLP, specifically designed to process sequences, including text.

Core Concepts

Turing Test

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A test designed by Alan Turing to assess a machine's ability to mimic human-like behavior and language. A machine passes if its outputs are indistinguishable from human responses."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A test designed by Alan Turing to assess a machine's ability to mimic human-like behavior and language. A machine passes if its outputs are indistinguishable from human responses.

Core Concepts

Unsupervised Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A learning approach where the model discovers hidden patterns in unlabeled data, similar to solving a puzzle without knowing the final picture."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A learning approach where the model discovers hidden patterns in unlabeled data, similar to solving a puzzle without knowing the final picture.

Techniques & Methods

Upstream Sampling

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A fine-tuning technique used in ChatGPT that generates multiple responses which are then ranked to select the best one."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A fine-tuning technique used in ChatGPT that generates multiple responses which are then ranked to select the best one.

Applications

User Interface (UI)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The interface facilitating human-computer interaction in various applications or devices."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The interface facilitating human-computer interaction in various applications or devices.

Techniques & Methods

Validation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"A process in machine learning where a model is assessed using a subset of data it has not seen during training to check its learning efficacy."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">A process in machine learning where a model is assessed using a subset of data it has not seen during training to check its learning efficacy.

Miscellaneous

Validation Data

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Data similar in structure to training data but used to assess a model's performance, particularly to detect overfitting."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Data similar in structure to training data but used to assess a model's performance, particularly to detect overfitting.

Core Concepts

Variance

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"The extent to which a machine learning model's intended function changes during the training phase."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">The extent to which a machine learning model's intended function changes during the training phase.

Techniques & Methods

Variation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"In natural language processing, a variation is also known as a query or utterance and works alongside intents to facilitate human-like conversation. A variation is a specific phrasing that someone might use to accomplish a certain action or intent. For example, if the intent is to \"pay by credit card,\" a variation could be, \"Can I use my card to pay?\""}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">In natural language processing, a variation is also known as a query or utterance and works alongside intents to facilitate human-like conversation. A variation is a specific phrasing that someone might use to accomplish a certain action or intent. For example, if the intent is to "pay by credit card," a variation could be, "Can I use my card to pay?"

Techniques & Methods

Vector Representation

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This is a method for converting words into a numerical format that computers can understand and process. Algorithms for vector representation are employed by ChatGPT to comprehend the semantics of words and produce suitable replies."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This is a method for converting words into a numerical format that computers can understand and process. Algorithms for vector representation are employed by ChatGPT to comprehend the semantics of words and produce suitable replies.

Miscellaneous

Vector Store

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This is a specialized data storage system designed for handling numerical representations of documents, known as embeddings. Optimized for efficient nearest-neighbor searches based on various distance metrics, vector stores serve as crucial components in the architecture of generative AI platforms."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This is a specialized data storage system designed for handling numerical representations of documents, known as embeddings. Optimized for efficient nearest-neighbor searches based on various distance metrics, vector stores serve as crucial components in the architecture of generative AI platforms.

Core Concepts

Weak AI

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Also referred to as narrow AI, weak AI specializes in a limited set of tasks and lacks the ability to learn or execute activities beyond its designated skill set. The majority of existing AI models fall into this category."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Also referred to as narrow AI, weak AI specializes in a limited set of tasks and lacks the ability to learn or execute activities beyond its designated skill set. The majority of existing AI models fall into this category.

Techniques & Methods

Word Embedding

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Similar to vector representation, word embedding is a technique for converting words into a format that computers can analyze. ChatGPT uses algorithms for word embedding to understand the semantics of text and provide fitting responses."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Similar to vector representation, word embedding is a technique for converting words into a format that computers can analyze. ChatGPT uses algorithms for word embedding to understand the semantics of text and provide fitting responses.

Miscellaneous

Yeoman's Work

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This term describes labor that is carried out willingly and without complaint, often consisting of tedious or strenuous tasks. ChatGPT can automate such tasks, like data entry or summarizing information, allowing humans to concentrate on more complex or creative endeavors."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This term describes labor that is carried out willingly and without complaint, often consisting of tedious or strenuous tasks. ChatGPT can automate such tasks, like data entry or summarizing information, allowing humans to concentrate on more complex or creative endeavors.

Techniques & Methods

Zero-Shot Learning

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"This denotes an AI model's capacity to comprehend and respond appropriately to tasks it was not explicitly trained on."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">This denotes an AI model's capacity to comprehend and respond appropriately to tasks it was not explicitly trained on.

Core Concepts

Zone of Proximal Development (ZPD)

<span data-sheets-root="1" data-sheets-value="{"1":2,"2":"Originally an educational theory, the Zone of Proximal Development refers to the range of tasks that a learner can perform with assistance. In machine learning, the concept is applied when training models on increasingly challenging tasks to enhance their learning capabilities."}" data-sheets-userformat="{"2":899,"3":{"1":0},"4":{"1":2,"2":16777215},"10":1,"11":4,"12":0}">Originally an educational theory, the Zone of Proximal Development refers to the range of tasks that a learner can perform with assistance. In machine learning, the concept is applied when training models on increasingly challenging tasks to enhance their learning capabilities.

AI Terms & Definitions

Terms