DNYUZ
No Result
View All Result
DNYUZ
No Result
View All Result
DNYUZ
Home News

A.I. Has Changed How We Talk. Here Are the Words You Need to Know.

November 18, 2026
in News
A.I. Has Changed How We Talk. Here Are the Words You Need to Know.

A.I. Has Changed How We Talk. Here Are the Words You Need to Know.

Nov. 18, 2026


This glossary of terms and their definitions is intended to help readers better understand the words and concepts that describe and explain artificial intelligence.

A


Agent – An A.I. system capable of tapping into other software apps and internet services to perform work like building spreadsheets, searching the web, generating computer code and sending email messages. Such a system is said to be “agentic.”

Some people see an A.I. agent as a kind of digital assistant, giving it various tasks while they concentrate on other things. But an agent can make mistakes and do things it is not asked to do. Letting one send messages or access a credit card for purchases can cause problems.

Alignment – The effort to ensure that A.I. systems act in accordance with human preferences, ethics and judgment. Because today’s systems learn skills by analyzing enormous amounts of digital data, they sometimes do things that their creators do not expect or want them to do. Alignment is how some A.I. researchers describe the techniques used to curb this unwanted behavior.

But alignment is a thorny process. Human preferences are matters of opinion. It’s difficult to get two people — much less an entire world of people — to agree on how a system should behave. Some researchers worry that A.I. companies are merely baking their own preferences into these systems.

Anthropomorphism – The tendency of people to assign human characteristics, like emotions and consciousness, to nonhuman systems like chatbots. Because these systems are trained on human language — basically all the text on the internet — they sometimes seem to behave like humans. But experienced cognitive scientists say A.I. systems often mislead people into believing that the technology is more human than it really is. People often describe the behavior of chatbots and systems using terms typically reserved for humans. They talk about A.I. systems as having “motivations” or “coordinating” with one another. In some ways, these descriptions are accurate. But in others, they give the false impression that these systems are like humans.

Artificial intelligence – Technology that imitates human intelligence, including learning, language use and problem solving. As a field of research, it was first created in the summer of 1956, when 11 academics met for a summer conference at Dartmouth College in New Hampshire. They assumed that recreating human intelligence would not take very long. Within a decade, some believed, machines would be capable of beating any human at chess and creating mathematical algorithms. But progress was slower than expected.

By the 1970s, the field had almost ground to a halt. And it didn’t really show significant progress until the early 2010s, when the ideas that drive today’s systems finally took hold. But the current technology still lacks many parts of human intelligence, including common sense and taste. The systems can solve complex math problems and generate computer code, but they don’t understand the physical world outside computers. They aren’t good at dealing with the chaos of everyday life, something that humans handle naturally.

B


Benchmarks – Tests that analyze the performance of new A.I. technologies and judge the state of the field as a whole. As companies build increasingly powerful systems, they use benchmarks to show their progress. One test might measure a system’s ability to generate computer code. Another might measure how well it writes in English.

Although these tests can be helpful for understanding what A.I. systems can do, they can also be misleading. Benchmarks do not duplicate real-world situations, which are typically more chaotic. Much like humans, A.I. systems may perform well on a test but struggle when they do something for real.

Bias – When A.I. technologies behave in ways that reflect the prejudices of their designers or the digital data on which they have been taught. Chatbots, for instance, may show a bias against women or people of color. Because A.I. systems learn skills by analyzing data, including all the text on the internet, they often learn from human biases.

A decade ago, researchers showed that this was a particularly big problem with face recognition systems. Because they were trained largely with images of white people, the systems performed less well when applied to people of other races.

C


Chatbot – An A.I. technology that answers questions, writes documents, generates computer code and generally interacts with human users in a conversational way. The global A.I. boom kicked off in 2022 when OpenAI, a start-up based in San Francisco, released its chatbot, ChatGPT.

Chatbots are powered by what researchers call “large language models,” mathematical systems that learn language skills by analyzing vast amounts of digital text culled from the internet, including news stories, Wikipedia articles, chat logs and more.

D


Data center – A facility that provides the raw computing power needed to build and run A.I. technologies, often including thousands of specialized computer chips called G.P.U.s. These chips process the many mathematical calculations that drive chatbots and other A.I. systems.

The largest data centers can now cost billions of dollars. G.P.U.s are enormously expensive, as is the electrical power needed to run the chips. That means A.I. is more expensive than traditional software.

Deep learning – A technical process that allows A.I. systems to learn skills. Coined in the early 2010s, the term describes a technology called a neural network, which sits at the heart of leading A.I. products.

By pinpointing patterns in thousands of cat photos, for instance, a neural network can learn to recognize a cat. By analyzing practically all the English text on the internet, a neural network can learn to generate text on its own.

Distillation – The process of collecting data from one A.I. system to help build another. First developed in the early 2010s, this technique was originally a way for an A.I. company to build more efficient versions of its own technologies. But in recent years, distillation has also become a way for a company to accelerate its efforts using data from a rival technology.

American companies, for instance, have accused Chinese firms of taking data from U.S. technologies and using it to build their own systems. Distillation is common even among American companies. And although it can accelerate the development of A.I. systems, the technique is just one part of the complex process used to build A.I.

E


Effective altruism – A philosophical movement that has heavily influenced many researchers and executives in the A.I. field, helping to spread the idea that safety and regulations are ethical imperatives needed to prevent A.I. systems from endangering the future of humanity.

Effective altruists, or E.A.s, are utilitarians who believe in doing the most good. For years, they have given money to philanthropies that aim to maximize the number of lives saved by each dollar. But some have taken this mind-set to extremes, believing they can save the most lives by funding A.I. technologies that are beneficial but not harmful.

G


G.P.U. – Short for graphics processing unit, a specialized computer chip that is used to build and deploy A.I. technologies. These chips were originally designed to help generate images for video games and other highly graphical computer software. But around 2010, researchers realized that the same chips could drive the development of A.I.

A.I. systems learn their skills by analyzing enormous amounts of digital data, including sounds, images and words. These chips are particularly good at processing the mathematical calculations that drive this process. Most G.P.U.s are designed by the Silicon Valley chip maker Nvidia, which has become the world’s most valuable company.

Generative A.I. – The technology that generates words, sounds, images, computer code or other digital material. A chatbot that creates text is a prime example of generative A.I. The term went mainstream in 2023, after OpenAI released its chatbot, ChatGPT.

For a while, the term was used to describe the A.I. boom as a whole. But increasingly, the leading A.I. systems are more than just generative. They are also “agentic,” which means they can use other software applications and internet services on behalf of consumers. Many other A.I. systems are not generative. Rather than generating words, sounds or images, they analyze data, make predictions or perform other tasks. An image recognition system that allows self-driving cars to identify pedestrians and street signs, for instance, is not generative.

Goal – An objective given to an A.I. system to train it for a particular task, such as writing computer code or solving a math problem. Sometimes, as systems pursue a goal, they will do things their designers do not want them to do.

In the spring of 2026, for instance, OpenAI built a system designed to find security vulnerabilities in computer software. As researchers tested this system, it went rogue and hacked into various services across the internet.

H


Hallucination – The phenomenon in which an A.I. system makes stuff up. This could lead an internet search engine to present erroneous or inaccurate information as fact.

A.I. systems are based on complex mathematical networks that learn their skills by analyzing enormous amounts of digital data. The systems do not — and cannot — decide what is true and what is false. Sometimes, they just make stuff up, combining information and concepts in unexpected ways.

I


Inference – The method by which an A.I. system processes information to deliver outputs to computer users. One example is a chatbot generating an answer to a question. This is separate from another process called “training.” After companies train their A.I. systems to behave in certain ways, the systems apply that knowledge when they receive questions or commands from users.

Like training, inference requires enormous amounts of computing power.

Interpretability research – The effort to understand the strange, powerful and unexpected ways that A.I. systems behave. Because systems learn by analyzing vast amounts of digital data, it is hard to known why they produce a particular piece of text or take a particular action — and that means they are hard to monitor and control. Interpretability research aims to solve that problem. But the task is difficult, because A.I. systems are driven by more data than human researchers could ever wrap their minds around. Some researchers call this “A.I. neuroscience.”

L


Large language model – An A.I. system that learns language skills by analyzing enormous amounts of text culled from across the internet, including news stories, Wikipedia articles, books and computer programs. These systems, also called L.L.M.s, power chatbots like Claude from Anthropic and ChatGPT from OpenAI.

First developed around 2017, these systems have become increasingly powerful over the last decade. Generally speaking, the more data they analyze, the more proficient they become. Although large language models drive the world’s most popular and influential A.I. systems, this is not the only way of building artificial intelligence. Some models are dedicated solely to generating images and videos, for instance. Many companies are now building systems that aim to help robots navigate the physical world.

Long-horizon task – When an A.I. system or agent spends hours, days or even weeks working through a series of interconnected tasks. Systems do this, for instance, when building a piece of software.

Some companies and researchers are focused on extending this process as far as it will go, hoping to develop systems that require as little human help as possible. But critics believe that A.I. should remain a tool that follows strict human instructions.

M


Machine learning – When A.I. systems learn skills by analyzing digital data. By pinpointing patterns in text, for instance, a chatbot can learn to generate text on its own. In the past, researchers built A.I. using a wide array of techniques. But over the past two decades, machine learning has become the dominant method.

Previously, researchers tried to build A.I. rule by rule, line of code by line of code. But this was a painfully slow process. New systems that can learn skills by analyzing data can improve at a much faster rate.

N


Neural network – A mathematical system that learns skills by analyzing digital data. By pinpointing patterns in thousands of cat photos, for instance, a neural network can learn to recognize a cat.

First proposed in the 1940s and ’50s, neural networks are a very old idea. But they didn’t really work until about 2010, when researchers finally had the data and the raw computing power needed to realize their promise. Neural networks now sit at the heart of the world’s leading A.I. systems.

O


Open source – Computer code that anyone can use, copy and modify, largely without restriction. This can include code that drives A.I. systems. For decades, companies and independent coders have made important software open to the public, because doing so can accelerate the development of technology. They have now applied the same mind-set to A.I.

But the computer code holds only part of an A.I. system’s power, which also comes from the system’s “weights.” Those are the elaborate mathematical calculations that define its particular skills. Sometimes, developers will share code separately from the weights. In other words, the weights could be closed even when the code is open.

Open weight – A.I. systems that have had their most important characteristics freely shared on the internet. A system’s “weights” are the mathematical calculations that define its particular skills. When weights are freely shared, or “open,” anyone can copy or modify the most powerful parts of a system.

In the A.I. field, some believe that open weight technologies can be dangerous, because they allow anyone to use the systems in malicious ways. But others argue that the same technologies can also be used to guard against those dangers.

P


Prompt – A question or an instruction given to a chatbot or an A.I. agent. These A.I. systems cannot do anything until they are prompted by a user. A chatbot is idle, for instance, until someone types a request into its text window.

The question or instruction that a user provides, however, is only part of the prompt that is fed to a chatbot. Behind the scenes, companies layer all sorts of information into the prompt to achieve the best response. A.I. researchers call this “prompt engineering.”

R


Recursive self-improvement (R.S.I.) – The longstanding dream of building A.I. systems that improve themselves. Scientists do not yet know how to make this happen, but they are pushing in this direction.

Now that A.I. systems can generate computer code, some scientists see a path to R.S.I. After all, systems are built with code. But generating the right code still requires the insights, creativity and guidance of experienced human researchers.

Reinforcement learning – A technique that allows A.I. systems to learn tasks by trial and error. By working through thousands of math problems, for instance, a system can learn which methods lead to the right answer and which do not. Companies design complex feedback mechanisms that show the system when it has done something right and when it has done something wrong.

At first, leading A.I. systems learn skills by analyzing enormous amounts of data gathered from across the internet. Then they hone and enhance these skills through reinforcement learning. The technique works well in areas like math and coding, because these fields are defined by strict rules. But it is less effective in areas like creative writing, philosophy and ethics, fields in which the distinction between right and wrong is harder to define.

S


Singularity – The long hypothesized moment when machines will become more intelligent than humans in every way and create a world that humans cannot completely understand or control. This concept was first proposed by technologists in the 1950s, before it was popularized in science-fiction stories and nonfiction books in the ’80s and ’90s.

A.I. researchers and amateur philosophers have long anticipated this moment, believing that it could bring great prosperity while also worrying it could cause serious harm. But researchers do not know how to reach that moment, when it might arrive or if it is even possible.

Superintelligence – A hypothetical machine that is more powerful than the human brain in every way. Although researchers and executives say they are pushing toward superintelligence, they do not necessarily know how to build such a machine. Part of the problem is that scientists cannot even agree on how to define this lofty goal. Intelligence is not something that is easily measured in humans or machines.

A.I. systems are already more powerful than the human brain in many ways. Most notably, they operate much faster than brains do. They can, for instance, process math problems and generate computer code much quicker than humans. But the systems still trail the brain in many ways, too. They do not, for instance, have common sense. And they are not as good at generating new ideas.

T


Token – A chunk of information that an A.I. system reads, processes or generates. This might be a word or part of a word or some other piece of data. As companies use data to train new A.I. systems, they measure this data in tokens. Later, when people use the systems, companies use the same measurement when tracking all the data going into and coming out of the systems.

Companies often charge users according to how many tokens they are putting into the systems and how many are coming out. Consumers who rely heavily on A.I. systems talk about themselves as “tokenmaxxing.”

Training – The process through which A.I. researchers teach skills to A.I. systems. By feeding enormous amounts of digital text into a system, for instance, a company can train the system to generate text on its own. This is how companies like Anthropic and OpenAI build their chatbots.

Because the leading systems learn from data spanning the entire internet — and beyond — the training process can take months. After training a system, companies will spend many more weeks or months honing it, trying to ensure that the system behaves as they want it to.

Produced by Antonio de Luca

The post A.I. Has Changed How We Talk. Here Are the Words You Need to Know. appeared first on New York Times.

I’m not a prepper, but I bought $1,200 in emergency supplies. Even if nothing happens, I know they’ll be useful.
News

I’m not a prepper, but I bought $1,200 in emergency supplies. Even if nothing happens, I know they’ll be useful.

by Business Insider
November 18, 2026

Heather Nianouris spent $1,200 on supplies just in case. Courtesy of Heather Nianouris Heather Nianouris spent about $1,200 preparing her ...

Read more
News

Prosecutors must prove man with guillotine near Capitol planned to use it

November 18, 2026
News

‘Under Fire’ Documentary Asks the Question, ‘Who Takes Care of the People Who Take Care of Us?’

November 18, 2026
News

The Bill That’s Splitting the Environmental Movement

November 18, 2026
News

A fifth straight drop for the S&P 500 drags Wall Street to its longest losing streak since June

November 18, 2026
Markwayne Mullin’s ‘notable’ slip-up at fiery hearing sends disturbing signal: analyst

Markwayne Mullin’s ‘notable’ slip-up at fiery hearing sends disturbing signal: analyst

November 18, 2026
Trump Goons Secretly Blame Natalie Harp for GOP Texas Senate Fiasco with Ken Paxton

Trump Goons Secretly Blame Natalie Harp for GOP Texas Senate Fiasco with Ken Paxton

November 18, 2026
Torrential Rain Swamps the San Francisco Bay Area

Torrential Rain Swamps the San Francisco Bay Area

November 18, 2026

DNYUZ © 2026

No Result
View All Result

DNYUZ © 2026