It’s almost impossible to ignore AI.
That doesn’t mean it’s always easy to understand. From agentic AI to UBI, tech CEOs, Wall Street, and politicians increasingly sound like they are speaking another language. The terms seem to change almost as quickly as AI models themselves advance.
Even if you don’t use AI, chances are your bank, your doctor, the streaming service you’re using, and maybe even your car do.
Here’s a list of the people, companies, and terms you need to know to talk about AI, in alphabetical order.
The AI terms you need to know
Agentic: A type of artificial intelligence that can make proactive, autonomous decisions with limited human input and is capable of operating autonomously around the clock. Popularized by services like OpenClaw, the proliferation of these tools is viewed as the biggest moment in generative AI since the release of ChatGPT.
AGI: “Artificial general intelligence,” or the ability of artificial intelligence to perform complex cognitive tasks such as displaying self-awareness and critical thinking, the way humans do. Crossing this theoretical threshold is the mission of many in the industry.
Alignment: A field of AI safety research that aims to ensure that the goals, decisions, and behaviors of AI systems are consistent with human values and intentions.
Bias: Because AI models are trained on data created by humans, they can also adopt the same fallible human biases present in that data. There are a number of different types of bias that AI models can succumb to, including prejudice bias, measurement bias, cognitive bias, and exclusion bias — all of which can distort the results.
Capability overhang: The term, credited to Microsoft Chief Technology Officer Kevin Scott, for the gap between what AI models are capable of and what real-world applications can currently utilize.
ChatGPT: OpenAI’s signature chatbot that launched to significant fanfare in 2022 and is often credited for kickstarting the AI race. GPT stands for Generative Pre-trained Transformer.
Claude: Anthropic’s flagship model was first launched in March 2023. While Anthropic’s core focus is on the enterprise business, Claude has been lauded for its ability to write code. In early 2026, Claude’s advancements caused sharp swings in some tech stocks.
Compute: The AI computing resources needed to train models and carry out tasks, including processing data. This can include GPUs, servers, and cloud services. As of 2026, many companies are adjusting their pricing, retiring old features, or limiting announcements due to resource constraints.
Context window: What a large language model can recall in response to a prompt or series of prompts. Generally speaking, increasing the amount of this “working memory” decreases the likelihood of hallucinations and enables users to have longer, more complex conversations with a model.
Data centers: Large warehouses filled with tens, if not hundreds, of thousands of advanced computer chips and graphics processing units, used to handle large amounts of data, storage, and complex processing required to power AI models. Unlike older iterations, AI data centers require significantly more space and energy because of the widely held assumption that AI models learn best at a massive scale. Cities and states are increasingly considering restrictions on development amid rising opposition to AI, most notably in the US.
Deepfake: An AI-generated image, video, or voice meant to appear real that tends to be used to deceive viewers or listeners. This deceptive content has been used to create non-consensual pornography and extort people for money.
Distillation: The process of extracting the reasoning process and learned knowledge of a larger, existing AI model to a new, smaller AI model — essentially, copying an AI model to start your own. US-based AI companies have accused their competitors, predominantly in China, of using large-scale attacks to advance their models.
Doomer: A derisive term for AI skeptics who express reservations about either the potential risks of AI development (developing technology that could turn against humanity) or even just pessimism that AI will achieve lofty ambitions like creating models capable of human-like reasoning.
Effective altruists: Broadly speaking, this is a social movement that stakes its claim in the idea that all lives are equally valuable and those with resources should allocate them to helping as many as possible. In the context of AI, effective altruists, or EAs, are interested in how AI can be safely deployed to reduce suffering caused by social ills such as climate change and poverty. Many AI leaders have distanced themselves from the movement in the wake of Sam Bankman-Fried’s fall, which included a 25-year prison sentence. (See also: e/accs and decels).
Federal preemption: The debate over whether each state should set some of its own AI-related policies, or if the federal government should set one uniform standard. President Donald Trump and his administration have unsuccessfully pushed for a moratorium on state-level AI laws, a policy that splits Republicans and tech companies.
Frontier models: Refers to the most advanced examples of AI technology. The Frontier Model Forum — an industry nonprofit launched by Microsoft, Google, OpenAI, and Anthropic in 2023 — defines frontier models as “large-scale machine-learning models that exceed the capabilities currently present in the most advanced existing models, and can perform a wide variety of tasks.”
Gemini: Google’s flagship AI model, first launched in 2023 under its former name “Bard.” As of late 2025, leading voices in the industry saw Gemini 3 as meeting, if not surpassing, ChatGPT’s capabilities. Google has said Gemini’s name was inspired by the zodiac constellation and NASA’s famed 1965 to 1968 project which helped form the foundation for putting humans on the Moon.
Gigawatts: A large unit of energy; a single gigawatt can power roughly 750,000 homes and is often used to put the enormous size of data center buildouts in perspective. In terms of computing power, 10 gigawatts is equal to roughly 4 to 5 million graphics processing units.
GPU: A computer chip, short for graphics processing unit, that companies use to train and deploy their AI models. Nvidia’s GPUs are used by Microsoft and Meta to run their AI models.
Hallucinations: A phenomenon where a large language model (see below) generates inaccurate information that it presents as a fact. For example, during an early demo, Google’s AI chatbot Bard hallucinated by generating a factual error about the James Webb Space Telescope.
Large language model: A complex computer program designed to understand and generate human-like text. The model is trained on large amounts of data and produces answers by scraping information across the web. Examples of LLMs include OpenAI’s GPT-5, Anthropic’s Claude Opus 4.7, and Google’s Gemini.
Machine learning: Refers to AI systems that can adapt and learn on their own, without following human instructions or explicit programming.
Multimodal: The ability for AI models to process text, images, and audio to generate an output. Users of ChatGPT, for instance, can now write, speak, and upload images to the AI chatbot.
Natural language processing: The umbrella term encompasses a variety of methods for interpreting and understanding human language. LLMs are one tool for interpreting language within the field of NLP.
Neural network: A machine learning program designed to think and learn like a human brain. Facial recognition systems, for instance, are designed using neural networks in order to identify a person by analyzing their facial features.
Open-source: A trait used to describe a computer program that anyone can freely access, use, and modify without asking for permission. Some AI experts have called for models behind AI, like ChatGPT, to be open-source so the public knows exactly how they are trained.
Optical character recognition: OCR is a technology that can recognize text within images — like scanned documents, text in photos, and read-only PDFs — and extract it into a text-only format that machines can read.
Prompt engineering: The process of asking AI chatbots questions that can produce desired responses. As a profession, prompt engineers are experts in fine-tuning AI models.
Rationalists: People who believe that the most effective way to understand the world is through logic, reason, and scientific evidence. In AI, rationalists seek to answer questions like how AI can be smarter, how AI can solve complex problems, and how AI can better process information around risk.
Responsible scaling policies: Guidelines for AI developers to follow that are designed to mitigate safety risks and ensure the responsible development of AI systems. Anthropic invoked its safety policy when it decided in April 2026 not to widely release Claude Mythos in part because the model was too good at hacking.
Singularity: A hypothetical moment where artificial intelligence becomes so advanced that it surpasses human intelligence. Think of a science fiction scenario where an AI robot develops agency and takes over the world.
Slop: A derisive term referring to AI-generated images, videos, or text, often of lower perceived quality.
Token: The building blocks of LLMs (see above) are units of text or parts of a word, both in the input and output of the model. AI companies use tokens to measure and charge usage.
Tokenmaxxing: In spring 2026, the popularity of agentic AI (see above) led to a school of thought that companies and their employees should use AI as much as possible in order to increase productivity. As time went on, the thinking began to fall out of favor because token usage alone does not convey what tasks are being done.
Transformer: A type of neural network that is at the core of large language models like OpenAI’s GPT, which in turn powers chatbots like ChatGPT. In fact, the last T in “GPT” is for transformer. Critically for AI, transformers were able to process massive datasets simultaneously, as opposed to earlier neural networks that processed data sequentially — dramatically reducing training time and enabling much larger models.
Universal basic income: A policy where the local, state, or federal government would guarantee a minimum income for citizens. Popularized by then-Democratic presidential hopeful Andrew Yang in 2020, the idea has taken on renewed relevance amid fears that AI may replace a significant number of jobs, potentially causing widespread unemployment. Alternatively, some, such as Elon Musk, believe AI could create an abundance for humanity, enabling everyone to become wealthy and achieve “universal high income.”
Vibe coding: Coined by OpenAI cofounder Andrej Karpathy in February 2025 to describe the feeling of freedom using generative AI coding models. Over time, the term became more commonly used to refer to how AI generative coding models can write code with minimal human input. Karpathy has since said that he prefers the term “agentic engineering” for such cases.
World models: A type of AI model that uses machine learning to understand physical properties, as opposed to LLMs (see above), which excel at language prediction but struggle with actual environments. Building these specific models is considered critical for the development of self-driving and ultimately AI robotics.
The top AI leaders
Sam Altman: The cofounder and CEO of OpenAI, the company behind ChatGPT. In 2023, Altman was ousted by OpenAI’s board before returning to the company as CEO days later. After his return, he has become one of the world’s most recognizable tech executives.
Dario Amodei: The CEO and cofounder of Anthropic, a major rival to OpenAI, where he previously worked. Amodei has frequently warned that world leaders are not preparing society for the scale of AI-related job displacement.
Demis Hassabis: The cofounder of DeepMind and now CEO of Google DeepMind, Hassabis leads its AI efforts at Alphabet.
Jensen Huang: The CEO and cofounder of Nvidia, the tech giant behind the specialized chips companies use to power their AI technology. Under his leadership, Nvidia has become the world’s largest company by market cap at over $5 trillion.
Alex Karp: The CEO and cofounder of Palantir, a defense and data company that has skyrocketed in value. Known as an iconoclastic leader, Karp called Palantir the “first to be anti-woke” and takes pride in the company’s national security business, especially its work with the US government.
Yann LeCun: Formerly Meta’s chief AI scientist, LeCun is a renowned researcher who is considered among the “Godfathers of AI” due to his work on deep learning with Nobel laureate Geoffrey Hinton and others. LeCun has been critical of some of Meta’s AI direction and is a leading skeptic of the extent to which Large Language Models (LLMs) will unlock the biggest breakthroughs in AI.
Elon Musk: The Tesla and SpaceX CEO founded artificial intelligence startup xAI in 2023 after a falling out with Altman over OpenAI’s shifting direction. In February 2026, SpaceX acquired xAI months before an IPO expected to be the largest in history.
Mira Murati: The CEO and cofounder of Thinking Machines, Murati has made waves in Silicon Valley since leaving OpenAI, where she was CTO and briefly interim CEO.
Satya Nadella: The CEO of Microsoft, the software giant behind the Bing AI-powered search engine Copilot, a suite of generative AI tools. Microsoft is also an investor in OpenAI.
Sundar Pichai: The CEO of Google, Pichai sustained some criticism of Google’s AI leadership after the release of OpenAI’s ChatGPT in 2022. By the end of 2025, some in the industry were beginning to proclaim Google had caught up, if not surpassed OpenAI.
Mustafa Suleyman: The cofounder of DeepMind, Google’s AI division, who left the company in 2022. He cofounded Inflection AI before he joined Microsoft as its chief of AI in March 2024.
Ilya Sutskever: The cofounder and chief scientist at Safe Superintelligence, Sutskever helped start OpenAI before eventually pushing for Altman’s ouster, a move he regrets. Like LeCun, Sutskever has expressed skepticism that scaling compute is enough to advance AI.
Alexandr Wang: Meta’s chief AI office has experienced a rapid rise since cofounding Scale AI in 2016, out of famed Silicon Valley startup incubator Y Combinator. In June 2025, Meta acquired a 49% stake in Scale AI and poached Wang as part of its campaign to lure top AI talent. Meta released Muse Spark in April 2026, the first LLM developed by Wang’s Superintelligence Labs team.
Liang Wenfeng: The hedge fund manager who founded Chinese AI startup DeepSeek in 2023. At the beginning of 2025, the startup made waves across the AI industry with its flagship model, R1, which reportedly rivals its top competitors in capability but operates at a fraction of the cost.
Mark Zuckerberg: The Facebook founder and Meta CEO who has been spending big to advance Meta’s AI capabilities, including training its own models and integrating the technology into its platforms.
Read the original article on Business Insider
The post A survival guide to the words behind the AI boom appeared first on Business Insider.
