Artificial intelligence is a deep and convoluted world. The scientists who work in this field often rely on jargon and lingo to explain what they’re working on. As a result, we frequently have to use those technical terms in our coverage of the artificial intelligence industry. That’s why we thought it would be helpful to put together a glossary with definitions of some of the most important words and phrases that we use in our articles. We will regularly update this glossary to add new entries as researchers continually uncover novel methods to push the frontier of artificial intelligence while identifying emerging safety risks. AGI Artificial general intelligence, or AGI, is a nebulous term. But it generally refers to AI that’s more capable than the average human at many, if not most, tasks. OpenAI CEO Sam Altman recently described AGI as the “equivalent of a median human that you could hire as a co-worker.” Meanwhile, OpenAI’s charter defines AGI as “highly autonomous systems that outperform humans at most economically valuable work.” Google DeepMind’s understanding differs slightly from these two definitions; the lab views AGI as “AI that’s at least as capable as humans at most cognitive tasks.” Confused? Not to worry — so are experts at the forefront of AI research. AI agent An AI agent refers to a tool that uses AI technologies to perform a series of tasks on your behalf — beyond what a more basic AI chatbot could do — such as filing expenses, booking tickets or a table at a restaurant, or even writing and maintaining code. However, as we’ve explained before, there are lots of moving pieces in this emergent space, so “AI agent” might mean different things to different people. Infrastructure is also still being built out to deliver on its envisaged capabilities. But the basic concept implies an autonomous system that may draw on multiple AI systems to carry out multistep tasks. Chain of thought Given a simple question, a human brain can answer without even thinking too much about it — things like “which animal is taller, a giraffe or a cat?” But in many cases, you often need a pen and paper to come up with the right answer because there are intermediary steps. For instance, if a farmer has chickens and cows, and together they have 40 heads and 120 legs, you might need to write down a simple equation to come up with the answer (20 chickens and 20 cows). In an AI context, chain-of-thought reasoning for large language models means breaking down a problem into smaller, intermediate steps to improve the quality of the end result. It usually takes longer to get an answer, but the answer is more likely to be correct, especially in a logic or coding context. Reasoning models are developed from traditional large language models and optimized for chain-of-thought thinking thanks to reinforcement learning. (See: Large language model) Techcrunch event Join us at TechCrunch Sessions: AI Secure your spot for our leading AI industry event with speakers from OpenAI, Anthropic, and Cohere. For a limited time, tickets are just $292 for an entire day of expert talks, workshops, and potent networking. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 REGISTER NOW Deep learning A subset of self-improving machine learning in which AI algorithms are designed with a multi-layered, artificial neural network (ANN) structure. This allows them to make more complex correlations compared to simpler machine learning-based systems, such as linear models or decision trees. The structure of deep learning algorithms draws inspiration from the interconnected pathways of neurons in the human brain. Deep learning AI models are able to identify important characteristics in data themselves, rather than requiring human engineers to define these features. The structure also supports algorithms that can learn from errors and, through a process of repetition and adjustment, improve their own outputs. However, deep learning systems require a lot of data points to yield good results (millions or more). They also typically take longer to train compared to simpler machine learning algorithms — so development costs tend to be higher. (See: Neural network) Diffusion Diffusion is the tech at the heart of many art-, music-, and text-generating AI models. Inspired by physics, diffusion systems slowly “destroy” the structure of data — e.g. photos, songs, and so on — by adding noise until there’s nothing left. In physics, diffusion is spontaneous and irreversible — sugar diffused in coffee can’t be restored to cube form. But diffusion systems in AI aim to learn a sort of “reverse diffusion” process to restore the destroyed data, gaining the ability to recover the data from noise. Distillation Distillation is a technique used to extract knowledge from a large AI model with a ‘teacher-student’ model. Developers send requests to a teacher model and record the outputs. Answers are sometimes compared with a dataset to see how accurate they are. These outputs are then used to train the student model, which is trained to approximate the teacher’s behavior. Distillation can be used to create a smaller, more efficient model based on a larger model with a minimal distillation loss. This is likely how OpenAI developed GPT-4 Turbo, a faster version of GPT-4. While all AI companies use distillation internally, it may have also been used by some AI companies to catch up with frontier models. Distillation from a competitor usually violates the terms of service of AI API and chat assistants. Fine-tuning This refers to the further training of an AI model to optimize performance for a more specific task or area than was previously a focal point of its training — typically by feeding in new, specialized (i.e., task-oriented) data.  Many AI startups are taking large language models as a starting point to build a commercial product but are vying to amp up utility for a target sector or task by supplementing earlier training cycles with fine-tuning based on their own domain-specific knowledge and expertise. (See: Large language model [LLM]) GAN A GAN, or Generative Adversarial Network, is a type of machine learning framework that underpins some important developments in generative AI when it comes to producing realistic data – including (but not only) deepfake tools. GANs involve the use of a pair of neural networks, one of which draws on its training data to generate an output that is passed to the other model to evaluate. This second, discriminator model thus plays the role of a classifier on the generator’s output – enabling it to improve over time.  The GAN structure is set up as a competition (hence “adversarial”) – with the two models essentially programmed to try to outdo each other: the generator is trying to get its output past the discriminator, while the discriminator is working to spot artificially generated data. This structured contest can optimize AI outputs to be more realistic without the need for additional human intervention. Though GANs work best for narrower applications (such as producing realistic photos or videos), rather than general purpose AI. Hallucination Hallucination is the AI industry’s preferred term for AI models making stuff up – literally generating information that is incorrect. Obviously, it’s a huge problem for AI quality.  Hallucinations produce GenAI outputs that can be misleading and could even lead to real-life risks — with potentially dangerous consequences (think of a health query that returns harmful medical advice). This is why most GenAI tools’ small print now warns users to verify AI-generated answers, even though such disclaimers are usually far less prominent than the information the tools dispense at the touch of a button. The problem of AIs fabricating information is thought to arise as a consequence of gaps in training data. For general purpose GenAI especially — also sometimes known as foundation models — this looks difficult to resolve. There is simply not enough data in existence to train AI models to comprehensively resolve all the questions we could possibly ask. TL;DR: we haven’t invented God (yet).  Hallucinations are contributing to a push towards increasingly specialized and/or vertical AI models — i.e. domain-specific AIs that require narrower expertise – as a way to reduce the likelihood of knowledge gaps and shrink disinformation risks. Inference Inference is the process of running an AI model. It’s setting a model loose to make predictions or draw conclusions from previously-seen data. To be clear, inference can’t happen without training; a model must learn patterns in a set of data before it can effectively extrapolate from this training data. Many types of hardware can perform inference, ranging from smartphone processors to beefy GPUs to custom-designed AI accelerators. But not all of them can run models equally well. Very large models would take ages to make predictions on, say, a laptop versus a cloud server with high-end AI chips. [See: Training] Large language model (LLM) Large language models, or LLMs, are the AI models used by popular AI assistants, such as ChatGPT, Claude, Google’s Gemini, Meta’s AI Llama, Microsoft Copilot, or Mistral’s Le Chat. When you chat with an AI assistant, you interact with a large language model that processes your request directly or with the help of different available tools, such as web browsing or code interpreters. AI assistants and LLMs can have different names. For instance, GPT is OpenAI’s large language model and ChatGPT is the AI assistant product. LLMs are deep neural networks made of billions of numerical parameters (or weights, see below) that learn the relationships between words and phrases and create a representation of language, a sort of multidimensional map of words. These models are created from encoding the patterns they find in billions of books, articles, and transcripts. When you prompt an LLM, the model generates the most likely pattern that fits the prompt. It then evaluates the most probable next word after the last one based on what was said before. Repeat, repeat, and repeat. (See: Neural network) Neural network A neural network refers to the multi-layered algorithmic structure that underpins deep learning — and, more broadly, the whole boom in generative AI tools following the emergence of large language models.  Although the idea of taking inspiration from the densely interconnected pathways of the human brain as a design structure for data processing algorithms dates all the way back to the 1940s, it was the much more recent rise of graphical processing hardware (GPUs) — via the video game industry — that really unlocked the power of this theory. These chips proved well suited to training algorithms with many more layers than was possible in earlier epochs — enabling neural network-based AI systems to achieve far better performance across many domains, including voice recognition, autonomous navigation, and drug discovery. (See: Large language model [LLM]) Training Developing machine learning AIs involves a process known as training. In simple terms, this refers to data being fed in in order that the model can learn from patterns and generate useful outputs. Things can get a bit philosophical at this point in the AI stack — since, pre-training, the mathematical structure that’s used as the starting point for developing a learning system is just a bunch of layers and random numbers. It’s only through training that the AI model really takes shape. Essentially, it’s the process of the system responding to characteristics in the data that enables it to adapt outputs towards a sought-for goal — whether that’s identifying images of cats or producing a haiku on demand. It’s important to note that not all AI requires training. Rules-based AIs that are programmed to follow manually predefined instructions — for example, such as linear chatbots — don’t need to undergo training. However, such AI systems are likely to be more constrained than (well-trained) self-learning systems. Still, training can be expensive because it requires lots of inputs — and, typically, the volumes of inputs required for such models have been trending upwards. Hybrid approaches can sometimes be used to shortcut model development and help manage costs. Such as doing data-driven fine-tuning of a rules-based AI — meaning development requires less data, compute, energy, and algorithmic complexity than if the developer had started building from scratch. [See: Inference] Transfer learning A technique where a previously trained AI model is used as the starting point for developing a new model for a different but typically related task – allowing knowledge gained in previous training cycles to be reapplied.  Transfer learning can drive efficiency savings by shortcutting model development. It can also be useful when data for the task that the model is being developed for is somewhat limited. But it’s important to note that the approach has limitations. Models that rely on transfer learning to gain generalized capabilities will likely require training on additional data in order to perform well in their domain of focus (See: Fine tuning) Weights Weights are core to AI training, as they determine how much importance (or weight) is given to different features (or input variables) in the data used for training the system — thereby shaping the AI model’s output.  Put another way, weights are numerical parameters that define what’s most salient in a dataset for the given training task. They achieve their function by applying multiplication to inputs. Model training typically begins with weights that are randomly assigned, but as the process unfolds, the weights adjust as the model seeks to arrive at an output that more closely matches the target. For example, an AI model for predicting housing prices that’s trained on historical real estate data for a target location could include weights for features such as the number of bedrooms and bathrooms, whether a property is detached or semi-detached, whether it has parking, a garage, and so on.  Ultimately, the weights the model attaches to each of these inputs reflect how much they influence the value of a property, based on the given dataset. Topics

The agentic AI explainer you need to impress your tech friends. Credit: Kilito Chan / Getty Images According to the AI overlords, this is the year of agentic AI.You may have seen Google announce its "agentic era" with a web browsing research assistant and an AI bot that calls nail salons and mechanics for you. OpenAI leadership talked about agentic AI being a "big theme in 2025" and has already introduced a research preview of Operator, an agent that can perform tasks on your behalf, and Deep Research, which "conducts multi-step research on the internet for complex tasks." Microsoft just unveiled Microsoft Discover, an enterprise agentic AI tool for scientists. And your next smartphone could have agentic features that can send custom messages, create calendar events, or pull together information from across different apps. If you've been nodding and smiling every time one of your tech friends mentions agentic AI, don't be embarrassed. This is a new entry in the AI glossary, but one that can no longer be ignored. You May Also Like So what exactly is agentic AI?"Agentic AI refers to a class of artificial intelligence systems designed to operate autonomously, perceive their environment, set goals, plan actions to achieve those goals, and execute those plans without continuous human intervention. These systems can learn and adapt over time based on feedback and new information." That's according to — what else? — Google's AI chatbot Gemini.Unlike generative AI, which is essentially a tool for creating some kind of output — code, text, audio, images, videos — agentic AI can autonomously perform tasks on a user's behalf. This is a step up from the standard AI chatbot experience. Instead of generating a response based on its training material, agentic AI can take additional steps, such as conducting internet searches and analyzing the results, consulting additional sources, or completing a task in another app or software.You may have heard this term used interchangeably with AI agents, but agentic AI is a broader term that encompasses technology that may not be fully autonomous but has some agent-like capabilities. So, OpenAI considers Operator an AI agent because it has contextual awareness and can perform tasks for you like sending text messages. And its Deep Research tool is agentic AI because it can autonomously crawl the web and compile a report for the user, though its capabilities pretty much stop there for now. Agentic AI is powered by more advanced reasoning models like ChatGPT o3 and Gemini 2.5 Pro Preview, which can break down complex tasks and make inferences. This brings large-language models like ChatGPT one step closer to mimicking how the human brain works. Unless you constantly retrain a generative AI model with new information, it can't learn new things, said Karen Panetta, IEEE Fellow and professor of engineering at Tufts University. "This other kind of AI can learn from seeing other examples, and it can be more autonomous in breaking down tasks and helping you with more goal-driven types of activities, versus more exploratory or giving back information."When combined with computer vision, which is what allows a model to "see" a user's computer screen, we get the agentic AI everyone is so excited about. Mashable Light Speed Want more out-of-this world tech, space and science stories? Sign up for Mashable's weekly Light Speed newsletter. By clicking Sign Me Up, you confirm you are 16+ and agree to our Terms of Use and Privacy Policy. Thanks for signing up! Why is everyone talking about agentic AI? Google's new AI shopping experience could utilize agentic AI to make purchases on your behalf. Credit: Google Agentic AI is not entirely new. Self-driving cars and robot vacuums could both be considered early examples of agentic AI. They're technologies with autonomous properties that rely on advanced sensors and cameras to make sense of their environment and react accordingly.But agentic AI is having its moment now for a few reasons. Crucially, the latest models have gotten better and more user-friendly (although sometimes too friendly). And as people begin to rely on AI chatbots like ChatGPT, there's a growing interest in using these tools to automate daily tasks like responding to emails. With agentic AI, you don't need to be a computer programmer to use ChatGPT for automation. You can simply tell the chatbot what to do in plain English and have it carry out your instructions. At least, that's the idea.Companies like OpenAI, Google, and Anthropic are banking on agentic AI because it has the potential to move the technology beyond the novelty chatbot experience. With agentic AI, tools like ChatGPT could become truly indispensable for businesses and individuals alike. Agentic AI tools could order groceries online, browse and buy the best-reviewed espresso machine for you, or even research and book vacations. In fact, Google is already taking steps in this direction with its new AI shopping experience.In the business world, companies are looking to agentic AI to resolve customer service inquiries and adjust stock trading strategies in real-time. What could possibly go wrong?Are there risks involved with unleashing autonomous bots in the wild? Why, yes. With an agent operating on your behalf, there's always a risk of it sending a sensitive email to the wrong person or accidentally making a huge purchase. And then there's the question of liability. "Am I going to be sued because I went and had my agent do something?" Panetta wondered. "Say I'm working as an officer of something, and I use an AI agent to make a decision, to help us do our planning, and then you lose that organization money."  Related Stories The major AI players have put safeguards in place to prevent AI agents from going rogue, such as requiring human supervision or approval for sensitive tasks. OpenAI says Operator won't take screenshots when it's in human override mode, and it doesn't currently allow its agent to make banking transactions. But what about when the technology becomes more commonplace? As we become more comfortable with agentic AI, will we become more passive and lax about oversight? Earlier in this article, we used Google Gemini to help define agentic AI. If we become dependent on AI tools for even simple learning, will human beings get dumber?Then there's the extensive data access we have to give agents. Sure, it would be convenient for ChatGPT to automatically filter, sort, or even delete emails. But do you want to give an AI company full access to every email you've ever sent or received?And what about bad actors that don't have such safeguards in place? Panetta warns of increasingly sophisticated cyberattacks utilizing agentic AI. "Because the access to powerful computing now is so cheap, that means that the bad actors have access to it," she said. "They can be running simulations and being able to come up with sophisticated schemes to break into your systems or connive you into taking out this equity loan." AI has always been a double-edged sword, with equally potent harms and benefits. And with agentic AI getting ready for primetime deployment, the stakes are getting higher.Disclosure: Ziff Davis, Mashable’s parent company, in April filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems. Cecily Mauran Tech Reporter Cecily is a tech reporter at Mashable who covers AI, Apple, and emerging tech trends. Before getting her master's degree at Columbia Journalism School, she spent several years working with startups and social impact businesses for Unreasonable Group and B Lab. Before that, she co-founded a startup consulting business for emerging entrepreneurial hubs in South America, Europe, and Asia. You can find her on X at @cecily_mauran.

"Artificial intelligence, if we're being frank, is a con: a bill of goods you are being sold to line someone's pockets."That is the heart of the argument that linguist Emily Bender and sociologist Alex Hanna make in their new book The AI Con. It's a useful guide for anyone whose life has intersected with technologies sold as artificial intelligence and anyone who's questioned their real usefulness, which is most of us. Bender is a professor at the University of Washington who was named one of Time magazine's most influential people in artificial intelligence, and Hanna is the director of research at the nonprofit Distributed AI Research Institute and a former member of the ethical AI team at Google.The explosion of ChatGPT in late 2022 kicked off a new hype cycle in AI. Hype, as the authors define it, is the "aggrandizement" of technology that you are convinced you need to buy or invest in "lest you miss out on entertainment or pleasure, monetary reward, return on investment, or market share." But it's not the first time, nor likely the last, that scholars, government leaders and regular people have been intrigued and worried by the idea of machine learning and AI.Bender and Hanna trace the roots of machine learning back to the 1950s, to when mathematician John McCarthy coined the term artificial intelligence. It was in an era when the United States was looking to fund projects that would help the country gain any kind of edge on the Soviets militarily, ideologically and technologically. "It didn't spring whole cloth out of Zeus's head or anything. This has a longer history," Hanna said in an interview with CNET. "It's certainly not the first hype cycle with, quote, unquote, AI."Today's hype cycle is propelled by the billions of dollars of venture capital investment into startups like OpenAI and the tech giants like Meta, Google and Microsoft pouring billions of dollars into AI research and development. The result is clear, with all the newest phones, laptops and software updates drenched in AI-washing. And there are no signs that AI research and development will slow down, thanks in part to a growing motivation to beat China in AI development. Not the first hype cycle indeed.Of course, generative AI in 2025 is much more advanced than the Eliza psychotherapy chatbot that first enraptured scientists in the 1970s. Today's business leaders and workers are inundated with hype, with a heavy dose of FOMO and seemingly complex but often misused jargon. Listening to tech leaders and AI enthusiasts, it might seem like AI will take your job to save your company money. But the authors argue that neither is wholly likely, which is one reason why it's important to recognize and break through the hype.So how do we recognize AI hype? These are a few telltale signs, according to Bender and Hanna, that we share below. The authors outline more questions to ask and strategies for AI hype busting in their book, which is out now in the US.Watch out for language that humanizes AIAnthropomorphizing, or the process of giving an inanimate object human-like characteristics or qualities, is a big part of building AI hype. An example of this kind of language can be found when AI companies say their chatbots can now "see" and "think."These can be useful comparisons when trying to describe the ability of new object-identifying AI programs or deep-reasoning AI models, but they can also be misleading. AI chatbots aren't capable of seeing of thinking because they don't have brains. Even the idea of neural nets, Hanna noted in our interview and in the book, is based on human understanding of neurons from the 1950s, not actually how neurons work, but it can fool us into believing there's a brain behind the machine.That belief is something we're predisposed to because of how we as humans process language. We're conditioned to imagine that there is a mind behind the text we see, even when we know it's generated by AI, Bender said. "We interpret language by developing a model in our minds of who the speaker was," Bender added.In these models, we use our knowledge of the person speaking to create meaning, not just using the meaning of the words they say. "So when we encounter synthetic text extruded from something like ChatGPT, we're going to do the same thing," Bender said. "And it is very hard to remind ourselves that the mind isn't there. It's just a construct that we have produced."The authors argue that part of why AI companies try to convince us their products are human-like is that this sets the foreground for them to convince us that AI can replace humans, whether it's at work or as creators. It's compelling for us to believe that AI could be the silver bullet fix to complicated problems in critical industries like health care and government services.But more often than not, the authors argue, AI isn't bring used to fix anything. AI is sold with the goal of efficiency, but AI services end up replacing qualified workers with black box machines that need copious amounts of babysitting from underpaid contract or gig workers. As Hanna put it in our interview, "AI is not going to take your job, but it will make your job shittier."Be dubious of the phrase 'super intelligence'If a human can't do something, you should be wary of claims that an AI can do it. "Superhuman intelligence, or super intelligence, is a very dangerous turn of phrase, insofar as it thinks that some technology is going to make humans superfluous," Hanna said. In "certain domains, like pattern matching at scale, computers are quite good at that. But if there's an idea that there's going to be a superhuman poem, or a superhuman notion of research or doing science, that is clear hype." Bender added, "And we don't talk about airplanes as superhuman flyers or rulers as superhuman measurers, it seems to be only in this AI space that that comes up."The idea of AI "super intelligence" comes up often when people talk about artificial general intelligence. Many CEOs struggle to define what exactly AGI is, but it's essentially AI's most advanced form, potentially capable of making decisions and handling complex tasks. There's still no evidence we're anywhere near a future enabled by AGI, but it's a popular buzzword.Many of these future-looking statements from AI leaders borrow tropes from science fiction. Both boosters and doomers — how Bender and Hanna describe AI enthusiasts and those worried about the potential for harm — rely on sci-fi scenarios. The boosters imagine an AI-powered futuristic society. The doomers bemoan a future where AI robots take over the world and wipe out humanity.The connecting thread, according to the authors, is an unshakable belief that AI is smarter than humans and inevitable. "One of the things that we see a lot in the discourse is this idea that the future is fixed, and it's just a question of how fast we get there," Bender said. "And then there's this claim that this particular technology is a step on that path, and it's all marketing. It is helpful to be able to see behind it."Part of why AI is so popular is that an autonomous functional AI assistant would mean AI companies are fulfilling their promises of world-changing innovation to their investors. Planning for that future — whether it's a utopia or dystopia — keeps investors looking forward as the companies burn through billions of dollars and admit they'll miss their carbon emission goals. For better or worse, life is not science fiction. Whenever you see someone claiming their AI product is straight out of a movie, it's a good sign to approach with skepticism. Ask what goes in and how outputs are evaluatedOne of the easiest ways to see through AI marketing fluff is to look and see whether the company is disclosing how it operates. Many AI companies won't tell you what content is used to train their models. But they usually disclose what the company does with your data and sometimes brag about how their models stack up against competitors. That's where you should start looking, typically in their privacy policies.One of the top complaints and concerns from creators is how AI models are trained. There are many lawsuits over alleged copyright infringement, and there are a lot of concerns over bias in AI chatbots and their capacity for harm. "If you wanted to create a system that is designed to move things forward rather than reproduce the oppressions of the past, you would have to start by curating your data," Bender said. Instead, AI companies are grabbing "everything that wasn't nailed down on the internet," Hanna said.If you're hearing about an AI product for the first time, one thing in particular to look out for is any kind of statistic that highlights its effectiveness. Like many other researchers, Bender and Hanna have called out that a finding with no citation is a red flag. "Anytime someone is selling you something but not giving you access to how it was evaluated, you are on thin ice," Bender said.It can be frustrating and disappointing when AI companies don't disclose certain information about how their AI products work and how they were developed. But recognizing those holes in their sales pitch can help deflate hype, even though it would be better to have the information. For more, check out our full ChatGPT glossary and how to turn off Apple Intelligence.