AI Explained: A video glossary of AI terms

Forget crypto and blockchain. ChatGPT is the new tech in town, and you need to know the lingo. Follow along as we boil down all the complex terms around AI, Large Language Models and Machine Learning into bite-sized videos. Soon enough, you'll have a comprehensive understanding of just how this technology is changing the game.

Enterprise AI Explained

Episode 53 | AI Term: Enterprise AI

Enterprise AI is a major catalyst reshaping businesses. Bringing together AI with human expertise to has the potential to transform operations, with AI seamlessly automating tasks, amplifying productivity, and liberating employees to focus on innovation and strategy.

Recent Episodes

Generative AI & LLMs Explained

Episode 01 | AI Term: Generative AI & LLMs

Have you ever wondered how generative AI is transforming app development and content creation? In this episode, we explore the power of large language models and how they're revolutionizing the way we generate new ideas, write stories, and even compose music. With generative AI, creating content has never been easier or more fluent. Tune in to discover how generative AI is changing the game.

Grounding Explained: How to stop AI hallucinations

Episode 02 | AI Term: Grounding

How can we prevent large language models from providing bad information? Well, today we’ll learn about a promising approach called grounding. Grounding has been shown to reduce the likelihood of errors, also known as hallucinations. So, if you want to stop hallucinations, watch to learn how to ground your large language model prompts.

Supervised vs. Unsupervised Learning Explained: What's the difference?

Episode 03 | AI Term: Supervised and Unsupervised Learning

Are you unsure about whether to use supervised or unsupervised learning for your AI project? In this video, you'll learn about the differences between these machine learning techniques and when to use each approach. While they may sound similar, the truth is that they can make a huge difference in how your AI models perform. Choosing the right approach for your AI project can be the key to unlocking its full potential.

Prompt Engineering Explained: Writing effective prompts is crucial.

Episode 04 | AI Term: Prompt Engineering

Are you tired of language models giving you unpredictable and irrelevant outputs? Then it's time to master the art of prompt engineering. We’ll provide you context and direction, so you can ensure that a language model generates the output you want. So why settle for generic responses when you can get precisely what you ask for?

Multimodal Language Models Explained: The next generation of LLMs

Episode 05 | AI Term: Multimodal Language Models

Multimodal language models are revolutionizing the way we interact with computers. This episode explores how these models are opening up a whole new world of possibilities beyond just language. From virtual assistants to automated customer service, these multimodals are set to transform the way we interact with technology, and usher in a new era of human-computer interaction.

Reinforcement Learning Explained: Correcting models with feedback

Episode 06 | AI Term: Reinforcement Learning

Reinforcement learning is transforming the field of AI. But what is it, and how does it work? Well, in this episode learn how this powerful combination of technology and human expertise has enabled ChatGPT to engage in natural and seamless conversations, and what it means for the future of AI.

 

Speech-to-text and Text-to-speech Explained

Episode 07 | AI Term: Speech-to-text, Text-to-speech

Discover the transformative power of Speech-to-Text and Text-to-Speech technologies, which are revolutionizing the way we communicate with machines. These innovations enable seamless interactions, turning spoken words into written text and vice versa, ultimately enhancing our communication experience with virtual assistants and chatbots.

Annotation Explained

Episode 08 | AI Term: Annotation

Behind every cutting-edge AI lies the crucial process of annotation. Expert annotators provide in-depth information to datasets, essentially creating a detailed guide that helps machine learning algorithms understand, learn, and deliver desired outputs in various scenarios. Watch this episode of AI Explained to learn how important annotation is for everything from ChatGPT to self-driving cars.

Chatbots vs. Conversational AI Explained

Episode 09 | AI Term: Chatbots vs. Conversational AI

Conversational AI is revolutionizing the way we interact with chatbots, allowing for a more natural human-like conversation. In this video we’ll discuss the differences between these two chat experiences and explore how conversational AI enables more efficient and intelligent communication across various applications.

Hallucination Explained

Episode 10 | AI Term: Hallucination

Hallucination sounds like a bad thing but really it’s a crucial aspect of generative AI, enabling imaginative creations. But there’s a trade off, in order to be imaginative you have to distort reality, and unfortunately  this can sometimes lead to incorrect or false information. Watch the full video to discover how engineers are working towards solving this issue and why it's essential to be cautious and not blindly trust AI-generated output.

The Cost of Large Language Models Explained

Episode 11 | AI Term: LLM Costs

Did you know that the cost of creating and maintaining a large language model can rival that of a Boeing 747? While developing such models can be expensive due to their size and complexity, smaller open-source smaller more affordable models also exist. Watch the full video to learn more about the costs of large language models!

Generative AI Explained

Episode 12 | AI Term: Generative AI

Did you know Generative AI can serve as your new creative partner and its core strengths promise to transform the way we interact with technology. Watch the full video to learn more about the process Generative AI follows to rapidly generate human-like content and how it’s changing the way the world works.

Probabilistic vs. Deterministic Explained

Episode 13 | AI Term: Probabilistic and Deterministic

To truly grasp the complexity of artificial intelligence it’s important to first understand the logic behind how it makes decisions. Some decisions are clear-cut, while others are more complicated with multiple possible outcomes, that’s where probabilistic and deterministic models come into play. Watch the full episode to learn how these decision models work and how they can be used to guide AI to solve problems.

Fine-tuning vs. Instruction-tuning Explained

Episode 14 | AI Term: Fine-tuning and Instruction-tuning

Tuning a machine learning model refers to the technique of tweaking the model in a way to produce a desired outcome. Often referred to as fine-tuning or instruction-tuning, these different techniques can be applied in different ways to produce wildly different outcomes. Watch the full video to learn more about how tuning can be used to optimize AI to perform specific tasks, or even better equip it to adapt to its environment.

Controllability Explained

Episode 15 | AI Term: Controllability

AI has the potential to transform many aspects of our lives, but it's not always perfect and can sometimes make mistakes with significant consequences. By employing techniques like interpretability, we can better understand and control AI decision-making, ensuring that it remains accurate, safe, and ethical in its applications. Watch the full episode to learn how engineers are deploying these methods get more control over AI

Artificial General Intelligence Explained

Episode 16 | AI Topic: Artificial General Intelligence

Artificial General Intelligence (AGI) has the power to unlock new discoveries and tackle some of the world's most daunting challenges. But it’s significantly more complex than narrow AI. Watch the full video to learn how engineers are integrating multiple AI techniques to build these AI systems that continuously improve and can potentially make the world a better place.

Stacking Explained

Episode 17 | AI Term: Stacking

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. Watch the full video for the full story.

Reasoning Explained

Episode 18 | AI Term: Reasoning

AI reasoning is a cognitive process that enables systems to solve problems, think critically, and generate new knowledge, similar to the human brain. Watch this video to learn how this technology drives advancements across diverse fields, such as autonomous vehicles and supply chain management, and can even assist in everyday decision-making.

Associative Memory Explained

Episode 19 | AI Term: Associative Memory

AI systems utilize associative memory, similar to the human brain, to store, retrieve, and process connected information for decision-making. By accessing real-time data, AI-powered tools like GPS devices can improve upon our abilities, providing up-to-date and optimal route suggestions. Watch the video to get deeper.

Foundation Models Explained

Episode 20 | AI Topic: Foundation Models

Foundation models are a broad category of AI models, encompassing large language models and others, such as computer vision and reinforcement learning models. Serving as a base for various applications, foundation models can be fine-tuned and combined with other AI models to cater to specific tasks and domains. Watch on to learn more!

Knowledge Generation Explained

Episode 21 | AI Topic: Knowledge Generation

Knowledge generation involves AI systems analyzing extensive data sets to discover patterns and insights, ultimately transforming raw facts into accessible information. This AI-driven process, which continuously refines and updates knowledge base content, saves time, boosts satisfaction, and adapts to changing customer or employee needs.

Collective Learning Explained

Episode 22 | AI Term: Collective Learning

Watch this video to learn about collective learning: a technique that identifies linguistic patterns by analyzing commonalities in large datasets. By recognizing shared aspects and understanding subtle language nuances, this approach can streamline processes such as IT support and be applied to domains like healthcare, education, and transportation.

OpenAI's Whisper Model Explained

Episode 23 | AI Topic: OpenAI's Whisper Model

OpenAI's Whisper is an innovative Automatic Speech Recognition system that translates spoken language into written text. Its applications range from capturing meeting details and language learning to medical transcription and smart home automation, significantly enhancing communication and accessibility between humans and machines.

LLM Benchmarking Explained

Episode 24 | AI Topic: LLM Benchmarking

Selecting the right large language model for specific enterprise applications entails understanding its capabilities and limitations, and conducting benchmark tests that simulate real-world scenarios. Watch this video to learn how this process allows for assessing criteria such as relevance, domain expertise, and data sensitivity to identify the ideal AI solution for your unique needs.

Zero-to-one Problem Explained

Episode 25 | AI Topic: Zero-to-one problem

The "zero-to-one problem" highlights the challenge of finding the initial breakthrough or solution in AI, which is often the hardest step. Once this hurdle is overcome, subsequent progress becomes much easier, paving the way for synergistic partnerships between AI and humans, fostering greater efficiency and innovation.

Voice Processing Explained

Episode 26 | AI Term: Voice Processing

AI systems convert speech to text and back to speech to improve efficiency, save computational resources, and seamlessly integrate with text-based applications. This process, used by popular voice assistants like Siri and Alexa, allows for better understanding of spoken requests and accurate, controlled audible responses.

Low-Code Explained

Episode 27 | AI Topic: Low-Code

Low-code platforms enable individuals with little or no programming expertise to design websites and apps easily using visual interfaces, pre-built components, and conversational AI. These user-friendly tools simplify the development process, empowering people from diverse backgrounds to bring their ideas to life without extensive coding knowledge.

Recursive Prompting Explained

Episode 28 | AI Topic: Recursive Prompting

Recursive prompting is an effective method to enhance AI models like OpenAI's GPT-3, guiding them to deliver more accurate and contextually relevant output. This strategy involves using a series of iterative prompts that build on previous responses, refining the AI's understanding and addressing ambiguities to provide better results in a conversational manner.

Responsible AI Explained

Episode 29 | AI Topic: Responsible AI

Responsible AI focuses on developing and deploying fair, unbiased, transparent, and socially beneficial artificial intelligence solutions. To avoid perpetuating implicit biases and ethical pitfalls, AI systems must be trained on diverse and unbiased data while preserving privacy. By adopting responsible AI principles, AI can enhance human decision-making and contribute positively to individual lives and the global community.

Generative Adversarial Networks Explained

Episode 30 | AI Topic: Generative Adversarial Networks

Generative Adversarial Networks (GANs) are a powerful type of AI neural network that create new, realistic data mimicking training data through the competition between a Generator and a Discriminator. With numerous applications such as imitating handwriting, composing music, and transforming images, GANs can generate astonishingly realistic images and content that are nearly indistinguishable from reality.

Parameter-Efficient Fine-Tuning Explained

Episode 31 | AI Term: Parameter-Efficient Fine-Tuning

Parameter-Efficient Fine-Tuning (PEFT) is an approach that optimizes resource usage while fine-tuning large AI models, such as GPT, T5, and BERT. Through techniques like Adapters, Low-Rank Adaptation, Prefix Tuning, and Prompt Tuning, PEFT adjusts a small number of key parameters and enables the creation of lightweight "tiny checkpoints." The result is a more efficient and accessible fine-tuning process, allowing AI advancements to remain resource-efficient and available to all.

Natural Language Ambiguity Explained

Episode 32 | AI Term: Natural Language Ambiguity

To build AI systems capable of tackling natural language ambiguity, we can use strategies like context analysis, probabilistic modeling, knowledge graphs, and annotation. These methods help AI understand multiple meanings, recognize patterns and cues, decipher complex language structures, and learn the intricacies of human language, allowing AI to work together with humans to interpret ambiguous expressions.

Data Augmentation Explained

Episode 33 | AI Term: Data Augmentation

Data Augmentation is a versatile technique that expands and diversifies training sets by modifying existing data, enhancing model performance, accuracy, and preventing overfitting. Applicable to images, audio, video, and text data, it transforms limited datasets into robust resources, ensuring AI models excel in various scenarios. Advanced techniques, like Generative Adversarial Networks, address data-intensive applications across industries, making Data Augmentation a vital solution for overcoming common AI challenges.

Adapters Explained

Episode 34 | AI Term: Adapters

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. Adapters showcase impressive potential for creating more versatile, agile, and adaptive AI models across various scenarios.

Multi-hop Reasoning Explained

Episode 35 | AI Term: Multi-hop reasoning

Multi-hop reasoning is a process enabling AI models to answer complex questions by connecting multiple pieces of information from various sources. Essential in natural language processing and machine reading comprehension tasks, it allows AI to comprehend relationships and context distributed across different parts of a text. As a result, multi-hop reasoning enhances the capabilities of conversational AI agents and other NLP-based applications, serving as a foundation for creating AI systems that can understand human language intricacies.

AI Copilots Explained

Episode 36 | AI Term: AI Copilots

AI copilots are powerful conversational interfaces that provide real-time support, guidance, and assistance across various tasks and decision-making processes. By processing and analyzing immense amounts of data, AI copilots help streamline tasks, enhance productivity, and simplify daily challenges, ultimately changing the way we interact with technology.

Overfitting Explained

Episode 37 | AI Term: Overfitting

Overfitting occurs when an AI model memorizes its training data, focusing on irrelevant details and patterns, resulting in poor generalization to new, unseen examples. To avoid overfitting, strategies such as adjusting model complexity, introducing varied data, and using cross-validation can ensure the AI learns relevant patterns, enabling it to handle real-world challenges more effectively.

Steerability Explained

Episode 38 | AI Term:  Steerability

Steerability allows AI models, like ChatGPT, to adapt their behavior according to user inputs and constraints to achieve desired outcomes. This skill offers users greater control over AI-generated results. Using the "system" message, users can prescribe their AI's style and task. Through prompt guidance, the AI focuses on nurturing critical thinking and problem-solving abilities, while considering constraints like the tutor role and student needs. The adaptability provided by steerability enhances AI models' responsiveness and capacity to meet an organization's unique requirements.

Stable Diffusion Explained

Episode 39 | AI Term: Stable Diffusion

Stable diffusion is an AI model that generates strikingly realistic images from text prompts, opening new creative possibilities for artists and casual users alike. By leveraging diffusion models and large neural networks, this technology converts text prompts into detailed visuals, democratizing image generation without requiring extensive technical expertise.

Deep Learning Explained

Episode 40 | AI Term: Deep Learning

Deep learning is an AI technique that uses neural networks to analyze huge datasets and teach itself to perform complex tasks, like recognizing images and understanding language, without explicit programming. By mimicking human cognition, this technology can match and even surpass human capabilities on a growing range of tasks, enabling new breakthroughs in fields from computer vision to natural language processing.

Neural Network Explained

Episode 41 | AI Term: Neural Network

Neural networks are AI systems modeled after the human brain that use interconnected neurons to recognize patterns and learn from data without explicit programming. Inspired by biological neurons, these versatile machine learning models power today's most advanced AI capabilities, from computer vision to speech recognition, by mimicking human cognition in innovative ways.

Artificial Intelligence (AI) Explained

Episode 42 | AI Term: Artificial Intelligence (AI)

Artificial intelligence refers to software and systems that can perform tasks and make decisions without explicit human guidance, enabled by techniques like machine learning that allow AI to improve through experience. From recommending movies to driving cars, AI is beginning to transform many aspects of our lives in both assistive and groundbreaking ways, though current technology remains specialized and limited compared to human cognition.

Structured vs. Unstructured Data Explained

Episode 43 | AI Term: Structured vs. Unstructured Data

Structured data is organized information like tables or databases, with predefined categories and relationships. Unstructured data is messier and more free-form, like social media posts, emails, or images, without rigid formatting rules. Both data types provide useful insights, so businesses should collect and analyze structured transaction records along with unstructured customer feedback.

Model Chaining Explained

Episode 44 | AI Term: Model Chaining

Model chaining refers to linking multiple AI systems together by feeding the output of one model directly into the next, combining their strengths to accomplish complex tasks. Chaining specialized AIs into an interconnected sequence allows them to mimic multifaceted human intelligence in innovative ways, enabling new applications and capabilities not possible with single models alone.

Natural Language Generation Explained

Episode 45 | AI Term: Natural Language Generation

Natural language generation (NLG) refers to AI systems that can produce understandable written or spoken output through learning linguistic rules from large datasets, allowing coherent text and speech production for applications like summarization and dialogue. Though current NLG models seem to lack full comprehension, active research aims to mimic human-level capabilities by improving training techniques and model architectures.

Automation Explained

Episode 46 | AI Term: Automation

Intelligent automation refers to the use of AI, machine learning, and software bots to handle high-volume, repetitive tasks in the enterprise, freeing employees to focus on more strategic work that leverages human strengths like creativity and empathy. By combining automated efficiency with human insight, businesses can reduce costs while enabling more meaningful, value-adding work.

Natural Language Processing Explained

Episode 47 | AI Term: Natural Language Processing

Natural language processing (NLP) refers to the ability of AI systems to analyze, understand, and generate human language, enabling more effective human-computer interaction through techniques that handle the complexity and nuance of real-world speech and text. As research in NLP continues to advance, the technology is moving closer to seamless contextual language understanding between humans and machines.

Stochastic Parrots Explained

Episode 48 | AI Term: Stochastic Parrots

Stochastic parrots are large language models that can fluently generate human-like text but do not actually comprehend the meaning behind the words, showcasing both the impressive capabilities and inherent limitations of some modern AI systems in achieving true language understanding. As research continues, developing models that grasp abstract concepts and semantics remains an ongoing challenge on the path to more human-like artificial intelligence.

Machine Learning Explained

Episode 49 | AI Term: Machine Learning

Machine learning refers to the ability of AI systems to learn behaviors and insights from data without explicit programming, enabling algorithms to improve at tasks like prediction, classification, and pattern recognition through exposure to large datasets. By autonomously identifying complex patterns within data, machine learning techniques allow computers to carry out sophisticated processes not possible with traditional code.

Natural Language Understanding Explained

Episode 50 | AI Term: Natural Language Understanding

Natural language understanding (NLU) refers to the ability of AI systems to analyze the meaning and intent behind written and spoken language by utilizing contextual information, allowing for more natural interactions with technologies like chatbots that can interpret linguistic nuances. By moving beyond text processing to deeper comprehension, NLU enables more fluid, human-like communication between humans and machines.

ChatGPT Explained

Episode 51 | AI Term: ChatGPT

ChatGPT is a conversational AI system from OpenAI that can provide remarkably human-like responses on nearly any topic through a deep learning model trained on vast amounts of online data. With recent upgrades like image recognition and speech capabilities, it showcases rapid advances in natural language AI — answering questions, generating content, and even engaging in verbal conversations.

Discriminative Model Explained

Episode 52 | AI Term: Discriminative Model

Discriminative models are a class of artificial intelligence models designed for precise data analysis and classification tasks. These models excel in accurately categorizing and prioritizing information, making them essential for applications such as diagnostics, fraud detection, and content relevance rankings. Their focus on well-defined boundaries and logical information arrangement demonstrates their reliability in real-world scenarios, offering valuable insights and decision-making capabilities across various industries.

Upcoming Episodes

Watch: AI Explained topic, Generative pre-trained transformer

Generative Pre-trained Transformer Explained

Episode 54 | AI Term: Generative pre-trained transformer

Generative pre-trained transformers (GPT) empower enterprises by automating tasks like customer service responses, lead generation, and personalized marketing, previously requiring significant resources. These readily available models democratize access to advanced AI capabilities, marking a pivotal evolution in enterprise software where knowledge, language, and automation converge for increased efficiency and competitiveness.

Watch: AI Explained topic, AI Plugin

AI Plugin Explained

Episode 55 | AI Term: AI Plugin

Plugins expand the capabilities of AI like ChatGPT by connecting it to external tools and services via APIs, unlocking new possibilities for AI beyond surfacing information and allowing it to take actions and become more versatile as a digital assistant. These plugins offer real-time information access, ensuring more accurate and relevant responses. While they bring significant opportunities across industries, their management is crucial to mitigate potential risks, yet they promise an exciting frontier for AI.

Watch: AI Explained topic, GPT-3 and GPT-4

GPT-3 and GPT-4 Explained

Episode 56 | AI Term: GPT-3 and GPT-4

GPT-3, an AI language model from OpenAI, powers content recommendations and creative writing by mimicking human text generation. GPT-4 builds on this with increased data and parameters for smarter, safer, and more sophisticated text generation. These advancements mark the rapid evolution of AI, offering substantial potential for transforming online interactions and business applications.

Watch: AI Explained topic, Few-shot learning

Few-shot learning Explained

Episode 57 | AI Term: Few-shot learning

Few-shot learning is changing the game by enabling AI models to learn from just a few examples, rather than requiring extensive labeled datasets. This approach allows AI systems to rapidly extract insights and make accurate predictions with minimal data. In this way, few-shot learning democratizes access to powerful AI models, making them accessible to companies without extensive data or coding expertise.

Moveworks Live 2024

Join us at Moveworks.global to learn how forward-thinking leaders are leaning into AI and unlocking new levels of business agility.

Register Now