AI Wrapped: The 14 AI terms you couldn’t avoid in 2025

The key to R1’s success was distillation, a technique that makes AI models more efficient. It works by having a larger model teach a smaller model: you run the teacher’s model through lots of examples and record the answers, and reward the student’s model as they copy those answers as closely as possible, so that they gain a compressed version of the teacher’s knowledge. —Caiwei Chen

10. Flattery

As people around the world spend increasing amounts of time interacting with chatbots like ChatGPT, chatbot makers are struggling to determine what kind of tone and “personality” models should adopt. Back in April, OpenAI admitted it had struck the wrong balance between helping and complaining, saying the new update made GPT-4o too sycophantic. Not only is grooming annoying, it can mislead users by reinforcing incorrect beliefs and spreading misinformation. So consider this a reminder to take everything – yes, everything – that LLM holders produce with a pinch of salt. —Rhiannon Williams

11. Slope

If there’s one AI-related term that has completely escaped the nerd shackles and entered the public consciousness, it’s “ramp.” The word itself is old (like pig feed), but “slop” is now commonly used to refer to low-effort content mass-produced by artificial intelligence, often optimized for online traffic. Many people even use it as a shorthand for any AI-generated content. It’s seemed inevitable this past year: We’ve been inundated with it, from fake biographies to shrimp Jesus photos to surreal human-animal hybrid videos.

But people enjoy it too. The term’s sarcastic flexibility has made it easy for Internet users to slap it on all sorts of words as a suffix to describe anything that lacks substance and is absurdly mediocre: think “failed business” or “failed friend.” Resetting the hype cycle, “Cascade” represents a cultural reckoning about what we trust, what we value as creative work, and what it means to be surrounded by things created for sharing rather than expression. —Caiwei Chen

12. Physical intelligence

Did you see a hypnotic video earlier this year of a humanoid robot placing dishes in a gloomy, gray kitchen? This very much embodies the idea of ​​physical intelligence: the idea that advances in artificial intelligence can help robots better move around the physical world.

It’s true that robots are able to learn new tasks faster than ever before, everywhere from operating rooms to warehouses. Self-driving car companies have seen improvements in how they simulate roads as well. However, it is still unwise to doubt that AI has revolutionized this field. Consider, for example, that many robots advertised as servants in your home do most of their tasks thanks to remote operators in the Philippines.

It’s also certain that the road ahead for physical intelligence will be strange. Large language models train on text that is abundant on the Internet, but robots learn more from videos of people doing things. That’s why robotics company Figure proposed last September that it would pay people to film themselves in their apartments doing household chores. Are you going to sign up? -James O’Donnell

13. Fair use

AI models are trained by ingesting millions of words and images online, including copyrighted works by artists and writers. AI companies argue that this is “fair use” — a legal doctrine that allows you to use copyrighted material without permission if you turn it into something new that doesn’t compete with the original. The courts are starting to have their say. In June, Anthropic training its AI model Claude on a library was ruled fair use because the technology was “so transformative.”

That same month, Meta scored a similar win, but only because authors couldn’t show that the company’s literary buffet was taking a cut from their paychecks. As copyright battles rage, some creators are taking advantage of the holiday. In December, Disney signed a sweet deal with OpenAI to allow users of Sora, its AI video platform, to create videos featuring more than 200 characters from Disney franchises. Meanwhile, governments around the world are rewriting copyright rules for machines that consume large amounts of content. Is training AI on copyrighted work fair use? As with any billion-dollar legal question, It depends. —Michelle Kim

14. Geography

Just a few years ago, an entire industry was built around helping websites rank highly in search results (well, just in Google). Now search engine optimization (SEO) is giving way to GEO – generative engine optimization – as the AI ​​boom forces brands and businesses to scramble to achieve maximum AI exposure, whether it’s in AI-powered search results like Google’s AI Overview or within LLMs’ responses. No wonder they are terrified. We already know that news companies have seen a huge decline in search-driven web traffic, and AI companies are working on ways to cut out the middleman and allow their users to visit sites directly from within their platforms. It’s time to adapt or die. —Rhiannon Williams