An AI copilot is a conversational interface that uses large language models to support users in various tasks and decision-making processes across multiple domains within an enterprise environment.
Adapters are an advanced method for making pre-trained AI models adaptable to new tasks without complete retraining. These modules save time, money, and resources by efficiently repurposing existing models for different tasks in areas like natural language processing, computer vision, and robotics.
Annotation is the process of labeling data with additional information to help machine learning algorithms understand and learn.
Artificial General Intelligence (or "AGI")
Artificial General Intelligence (AGI) refers to an AI system that possesses a wide range of cognitive abilities, much like humans, enabling them to learn, reason, adapt to new situations, and devise creative solutions across various tasks and domains, rather than being limited to specific tasks as narrow AI systems are.
Artificial Intelligence (or “AI”)
The simulation of human intelligence in machines that are programmed to think and learn like humans. Example: A self-driving car that can navigate and make decisions on its own using AI technology.
Associative memory refers to a system's ability to store, retrieve, and process related information based on connections between elements, enabling it to efficiently identify and use relevant data for decision-making.
Automatic speech recognition (ASR) is a technology that transcribes spoken language into text.
Automation refers to the use of technology to perform tasks with minimal human intervention.
AI plugins are specialized software components that allow AI systems to interface with external applications and services.
Benchmarking is the process of evaluating and comparing products or systems using standardized tests to gauge performance and capabilities.
A chat interface built on top of GPT-3.5. GPT-3.5 is a large language model developed by OpenAI that is trained on a massive amount of internet text data and fine-tuned to perform a wide range of natural language tasks. Example: GPT-3.5 has been fine-tuned for tasks such as language translation, text summarization, and question answering.
Collective learning is an AI training approach that leverages diverse skills and knowledge across multiple models to achieve more powerful and robust intelligence.
Controllability is the ability to understand, regulate, and manage an AI system's decision-making process, ensuring its accuracy, safety, and ethical behavior, and minimizing the potential for undesired consequences.
A subfield of AI that focuses on developing systems that can understand and generate human-like language and conduct a back-and-forth conversation. Example: A chatbot that can understand and respond to customer inquiries in a natural and human-like manner.
A user-friendly interface that allows the user to ask questions and receive answers. Depending on the backend system that fuels the chatbot, it can be as basic as pre-written responses to a fully conversational AI that automates issue resolution.
The cost of large language models primarily stems from their size and complexity, which demand significant computational power, storage, and resources for training and deployment. These factors can result in substantial expenses for building, maintaining, and using such models, sometimes amounting to several dollars per conversation or thousands of dollars per month.
Data Augmentation is a technique used to artificially increase the size and diversity of a training set by creating modified copies of the existing data. It involves making minor changes such as flipping, resizing, or adjusting the brightness of images, to enhance the dataset and prevent models from overfitting.
A subfield of ML that uses neural networks with multiple layers to learn from data. Example: A deep learning model that can recognize objects in an image by processing the image through multiple layers of neural networks.
A deterministic model follows a specific set of rules and conditions to reach a definite outcome, operating on a cause-and-effect basis.
Discriminative models are algorithms designed to directly model and learn the boundary between different classes or categories in a dataset.
Enterprise AI refers to the strategic integration and deployment of AI within an organizational framework to enhance various business processes, decision-making, and overall operational efficiency.
Explainability refers to techniques that make AI model decisions and predictions interpretable and understandable to humans.
Extensibility in AI refers to the ability of AI systems to expand their capabilities to new domains, tasks, and datasets without needing full retraining or major architectural changes.
Extraction is the ability of generative models to analyze large datasets and extract relevant patterns, trends, and specific pieces of information.
Few-shot learning is a machine learning approach where models can learn concepts from just a few labeled examples, often 5 or less per category.
The process of adapting a pre-trained model to a specific task by training it on a smaller dataset. For example, an image classification model trained on all intersection pictures can be fine turned to detect when a car runs a red light. At Moveworks, we’ve been fine-tuning LLMs for enterprise support for years.
Foundation models are a broad category of AI models which include large language models and other types of models such as computer vision and reinforcement learning models. They are called "foundation" models because they serve as the base upon which applications can be built, catering to a wide range of domains and use cases.
GPT-3 is the 3rd version of the GPT-n series of models. It has 175 billion parameters — knobs that can be tuned — with weights to make predictions. Chat-GPT uses GPT-3.5, which is another iteration of this model.
GPT-4 is the latest model addition to OpenAI's deep learning efforts and is a significant milestone in scaling deep learning. GPT-4 is also the first of the GPT models that is a large multimodal model, meaning it accepts both image and text inputs and emits text outputs.
Generation is the ability of a generative model to create brand new, original content such as text, images, audio or video from scratch.
Generative AI models create new data by discovering patterns in data inputs or training data. For example, creating an original short story based on analyzing existing, published short stories.
Grounding is the process of anchoring artificial intelligence (AI) systems in real-world experiences, knowledge, or data. The objective is to improve the AI's understanding of the world, so it can effectively interpret and respond to user inputs, queries, and tasks. Grounding helps AI systems become more context-aware, allowing them to provide better, more relatable, and relevant responses or actions.
Generative adversarial networks (or "GANs")
GANs are a powerful type of neural network capable of generating new, never-seen-before data that closely resembles the training data.
Generative Pre-Trained Transformer
Generative pre-trained transformers (GPT) are neural network models trained on large datasets in an unsupervised manner to generate text.
Hallucination refers to a situation wherein an AI system, especially one dealing with natural language processing, generates outputs that may be irrelevant, nonsensical, or incorrect based on the input provided. This often occurs when the AI system is unsure of the context, relies too much on its training data, or lacks a proper understanding of the subject matter.
Instruction-tuning is an approach where a pre-trained model is adapted to perform specific tasks by providing a set of guidelines or directives that outline the desired operation.
Intelligence augmentation refers to empowering human capabilities through synergistic combinations of AI systems and traditional tools.
Interpretability refers to how inherently understandable or explainable an AI model is based on its architecture, logic, and behavior.
K-shot learning is a machine learning approach where models learn from only k labeled examples per class, where k is a small number like 1-5.
Knowledge generation involves training models on extensive datasets, allowing them to analyze data, discover patterns, and craft new insights.
Latency refers to the time delay between when an AI system receives an input and generates the corresponding output.
Low-code is a visual approach to software development that enables faster delivery of applications through minimal hand-coding.
Large Language Model (or “LLM”)
A type of deep learning model trained on a large dataset to perform natural language understanding and generation tasks. There are many famous LLMs like BERT, PaLM, GPT-2, GPT-3, GPT-3.5, and the groundbreaking GPT-4. All of these models vary in size (number of parameters that can be tuned), in the breadth of tasks (coding, chat, scientific, etc.), and in what they're trained on.
A subfield of AI that involves the development of algorithms and statistical models that enable machines to improve their performance with experience. Example: A machine learning algorithm that can predict which customers are most likely to churn based on their past behavior.
Model chaining is a technique in data science where multiple machine learning models are linked in a sequence to make predictions or analyzations.
Multi-hop is a term often used in natural language processing and, more specifically, machine reading comprehension tasks. It refers to the process by which an AI model retrieves answers to questions by connecting multiple pieces of information present in a given text or across various sources and systems, rather than directly extracting the information from a single passage.
Multimodal Language Models are a type of deep learning model trained on large datasets of both textual and non-textual data.
Zero/Single/Few shot learning are variations of the same concept – providing a model with little or no training data to classify new data and guide predictions. A “shot” represents a single training example. Fun fact: Within the GPT prompt, you can ask for “N” examples to improve the accuracy of the response.
Natural language ambiguity refers to situations where a word, phrase, or sentence can have multiple meanings, making it challenging for both humans and AI systems to interpret correctly.
Natural Language Generation (or “NLG”)
A subfield of AI that produces natural written or spoken language.
Natural Language Processing (or “NLP”)
A subfield of AI that involves programming computers to process massive volumes of language data. Focuses on transforming free-form text into a standardized structure.
Natural Language Understanding (or “NLU”)
A subtopic of NLP that analyzes text to glean semantic meaning from written language. That means understanding context, sentiment, intent, etc.
No-code is an approach to designing and using applications that doesn't require any coding or knowledge of programming languages.
A machine learning model inspired by the human brain's structure and function that's composed of layers of interconnected nodes or "neurons." Example: A neural network that can recognize handwritten digits with high accuracy.
The organization that developed ChatGPT. More broadly speaking, OpenAI is a research company that aims to develop and promote friendly AI responsibly. Example: OpenAI's GPT-3 model is one of the largest and most powerful language models available for natural language processing tasks.
The process of adjusting the parameters of a model to minimize a loss function that measures the difference between the model's predictions and the true values. Example: Optimizing a neural network's parameters using a gradient descent algorithm to minimize the error between the model's predictions and the true values.
A problem that occurs when a model is too complex, performing well on the training data but poorly on unseen data. Example: A model that has memorized the training data instead of learning general patterns and thus performs poorly on new data.
Parameter-efficient Fine-tuning (or "PEFT")
Parameter-Efficient Fine-Tuning, also known as PEFT, is an approach that helps you improve the performance of large AI models while optimizing for resources like time, energy, and computational power. To do this, PEFT focuses on adjusting a small number of key parameters while preserving most of the pretrained model's structure.
Training a model on a large dataset before fine-tuning it to a specific task. Example: Pre-training a language model like ChatGPT on a large corpus of text data before fine-tuning it for a specific natural language task such as language translation.
Identifying inputs — prompts — that result in meaningful outputs. As of now, prompt engineering is essential for LLMs. LLMs are a fusion of layers of algorithms and, consequently, have limited controllability with few opportunities to control and override behavior. An example of prompt engineering is providing a collection of templates and wizards to direct a copywriting application.
A probabilistic AI model makes decisions based on probabilities or likelihoods.
AI reasoning is the process by which artificial intelligence systems solve problems, think critically, and create new knowledge by analyzing and processing available information, allowing them to make well-informed decisions across various tasks and domains.
Recursive prompting is a strategy for guiding AI models like OpenAI's GPT-4 to produce higher-quality output. It involves providing the model with a series of prompts or questions that build upon previous responses, refining both the context and the AI's understanding to achieve the desired result.
A type of machine learning in which a model learns to make decisions by interacting with its environment and receiving feedback through rewards or penalties. GPT uses reinforcement learning from human feedback. When tuning GPT-3, human annotators provided examples of the desired model behavior and ranked outputs from the model.
Responsible AI refers to the approach of creating, implementing, and utilizing AI systems with a focus on positively impacting employees, businesses, customers, and society as a whole, ensuring ethical intentions and fostering trust, which in turn enables companies to confidently scale their AI solutions.
A subfield of NLP that focuses on modeling sequential data such as text, speech, or time series data. Example: A sequence model that can predict the next word in a sentence or generate coherent text.
The process of converting spoken words into written text.
Stable diffusion is an artificial intelligence system that uses deep learning to generate images from text prompts.
Stacking is a technique in AI that combines multiple algorithms to enhance overall performance. By blending the strengths of various AI models, stacking compensates for each model's weaknesses and achieves a more accurate and robust output in diverse applications, such as image recognition and natural language processing.
AI steerability refers to the ability to guide or control an AI system's behavior and output according to human intentions or specific objectives. This involves designing AI models with mechanisms that understand and adhere to the preferences provided by users, while avoiding unintended or undesirable outcomes. Improving steerability requires ongoing research and refinement, including techniques like fine-tuning, rule-based systems, and implementing additional human feedback loops during AI development.
Strong AI refers to machines possessing generalized intelligence and capabilities on par with human cognition.
Structured data refers to information that is organized and labeled in a standardized format.
Summarization is the ability of generative models to analyze large texts and produce concise, condensed versions that accurately convey the core meaning and key points.
A type of machine learning in which a model is trained on labeled data to make predictions about new, unseen data. Example: A supervised learning algorithm that can classify images of handwritten digits based on labeled training data.
Stochastic parrots are AI systems that use statistics to convincingly generate human-like text, while lacking true semantic understanding behind the word patterns.
Text-to-speech (TTS) is a technology that converts written text into spoken voice output. It allows users to hear written content being read aloud, typically using synthesized speech.
The process of breaking text into individual words or subwords to input them into a language model. Example: Tokenizing a sentence "I am ChatGPT" into the words: “I,” “am,” “Chat,” “G,” and “PT.”
A type of neural network architecture designed to process sequential data, such as text. Example: The transformer architecture is used in models like ChatGPT for natural language processing tasks.
Unstructured data is any information that isn't arranged in a pre-defined model or structure, making it tough to collect, process, and analyze.
A type of machine learning in which a model is trained on unlabeled data to find patterns or features in the data. Example: An unsupervised learning algorithm that can cluster similar images of handwritten digits based on their visual features.
Voice processing in AI refers to the pipeline of speech-to-text conversion followed by text-to-speech synthesis.
OpenAI’s Whisper is an AI system developed to perform automatic speech recognition (ASR), the task of transcribing spoken language into text.
Weak AI refers to narrow systems that excel at specific tasks within limited contexts, but lack generalized intelligence and adaptability outside their domain.
Weak-to-strong generalization is an AI training approach that uses less capable models to guide and constrain more powerful ones towards better generalization beyond their narrow training data.
Zero-shot learning is a technique in which a machine learning model can recognize and classify new concepts without any labeled examples.
The zero-to-one problem refers to the difficulty of finding an initial solution when addressing complex challenges, which is often disproportionately challenging compared to subsequent progress.