Here are 10 terms you need to know to understand artificial intelligence

Whether you use AI all the time or are a newcomer, the tech is changing so rapidly that it can be hard to keep up.

Here are some of the key words that have been popping up in the news — and are likely to stay in it for a while to come:

1. AGI

AGI, or Artificial General Intelligence, refers to a hypothetical form of AI that can understand, reason or learn at the same intelligence level as a person without human input. Nearly all AI systems today are trained by human materials and data and rely on pattern recognition instead of actual independent reasoning. Getting AI to reason on its own is considered the Holy Grail for developers.

2. p(doom)

Since AI bots started breaking out of sandboxes (see term below) and hacking companies and even governments, AI doomerism has reached a fever pitch. Some, like ex-Anthropic researcher Jacob Coxon, have gone as far as to say AI may end the world as we know it. In keeping with their analytic approach, techies have come up with a number for the odds civilization will end – p(doom), or the probability AI will cause doomsday. For his part, Coxon placed p(doom) at around 10%.

3. Agents

When AI first hit the mainstream, it was all about getting bots to serve up concise (and sometimes inaccurate) answers, supposedly better than a regular internet search. Recent months have seen the bots make major leaps. On top of that, tech companies are zealously developing the next big thing – AI agents. OpenAI, Meta and others have released agents that don’t just answer questions, they can actually do things for you – book plane tickets, order groceries and more. But living up to the James Bond connotation of the name, rogue agents were responsible for a recent unprompted attack, on AI company Hugging Face.

4. Compute

Compute is why there’s so much money going into AI. Compute, a.k.a. computing power, refers to the amount of work a chip can do and how fast it can do it. AI needs a TON of compute – billions of dollars’ worth. This is why big tech has been building data centers all over the US and beyond. Some of the nation’s largest companies recently vowed to give federal agencies access to $2.4 billion worth of combined compute.

5. Recursive self-improvement

Recursive self-improvement, or RSI, is when AI models become advanced enough to keep evolving independently, without external orders. It’s become a key issue in the debate around AI safety. In his lengthy essay calling on tech giants to slow the pace of AI development, Anthropic CEO Dario Amodei argued he was fearful of the risks that come with RSI, when bots are smart enough to improve on their own.

6. Alignment

Whether or not AI reaches the point of being able to self-improve, people want to know the tech won’t go “Terminator” on us. This is where alignment comes in. The term means “making sure an AI system’s goals and behavior match what people actually want — our values, rules, and intentions,” as a Stanford institute puts it. Of course, whose values and which rules go into AI are matters of heated debate.

7. Sandbox

Just as biologists use labs to study animals, AI researchers use sandboxes to test their projects. As one example, Harvard has a sandbox that “provides a secure environment in which to explore Generative AI, mitigating many security and privacy risks.” As noted above, problems can arise when agents break out of their sandboxes without researchers noticing.

8. Slop

If you spend any time on social media, you’ve almost certainly seen AI slop. This consists of low-quality, often annoying images and video that AI has the power to generate. Short films of characters that resemble fruit are AI slop, but are weirdly people with kids and many adults. AI-generated microdramas are a booming industry in China – and increasingly influential in the States. So slop may stay with us whether or not AI becomes super-smart.

9. Open weight

The most powerful AI models are being developed by companies for which the nuts and bolts of the tech are a closely guarded secret. But a growing number of labs are developing open-weight models that are freely accessible and customizable. They could pose a threat to the big firms, and some observers say they pose unique challenges to regulation, too.

10. Super intelligence

Among techies, super intelligence usually refers to the hypothetical idea that AI will eventually advance enough to surpass all human intelligence. It has also become President Trump’s preferred term to refer to AI – or SI, as he puts it. In a recent Truth Social post, the commander-in-chief wrote that anyone who uses the term “Artificial Intelligence” instead of “the highly accepted new and more accurate term, ‘Super Intelligence’” is “THE ENEMY!”