The key to R1’s success was distillation, a technique that makes AI models more efficient. It works by having a larger model frame a smaller model: you run the teacher model over many examples and record the answers, and reward the student model when it copies those answers as closely as possible, so it gets a compressed version of the teacher’s knowledge. —Caiwei Chen
10. Sycophany
As people around the world spend more and more time interact with chatbots Like ChatGPT, chatbot creators struggle to find the type of tone and “personality” models should adopt. Last April, OpenAI admitted to striking a poor balance between helping and whining, saying a new update had made it GPT-4o. too sycophantic. The fact that it harms you is not only irritating: it can mislead users by reinforcing their incorrect beliefs and spreading misinformation. So consider this your reminder to take everything – yes, everything – that LLMs produce with a pinch of salt. —Rhiannon Williams
11. Slope
If there’s one AI-related term that has completely escaped the nerd pen and entered the public consciousness, it’s “slop.” The word itself is old (think pig feeding), but “slop” is now commonly used to refer to mass-produced, effortless AI-generated content, often optimized for online traffic. Many people even use it as a shorthand for any AI-generated content. This seemed essential to us last year: we have been marinating in it since fake biographies has Jesus shrimp pictures to surreal human-animal hybrid videos.
But people also have fun with it. The term’s sardonic flexibility has made it easy for people to use it with all sorts of words as a suffix to describe anything that lacks substance and is absurdly mediocre: think “work slop” or “friend slop.” As the hype cycle resets, “slop” marks a cultural reckoning about what we trust, what we value as creative work, and what it means to be surrounded by things designed for engagement rather than expression. —Caiwei Chen
12. Physical intelligence
Have you encountered hypnosis video from the beginning of the year, of a humanoid robot putting away dishes in a dark grayscale kitchen? This roughly embodies the idea of physical intelligence: the idea that advances in AI can help robots move better in the physical world.
It’s true that robots were able to learn new tasks faster than ever, everywhere since operating rooms has warehouses. Self-driving car makers have seen improvements in the way they operate. simulate the roads too. That said, it is still wise to be skeptical that AI has revolutionized the field. Consider, for example, that many robots advertised as butlers in your home perform the majority of their tasks through remote operators in the Philippines.
The road ahead for physical intelligence will also certainly be strange. Large language models train on text, which is abundant on the Internet, but robots learn more from videos of people doing things. That’s why robot company Figure suggested in September that it pay people to film themselves in their apartment doing chores. Would you like to register? —James O’Donnell
13. Fair use
AI models are trained by devouring millions of words and images from the Internet, including copyrighted works of 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 weigh in. In June, Anthropic’s training of its AI model Claude on a library of books was deemed fair because the technology was “extremely transformative.”
The same month, Meta scored a similar victorybut only because the authors could not demonstrate that the company’s literary buffet reduced their salaries. As copyright battles mount, some creators are cashing in on the party. In December, Disney signed a brilliant affair with OpenAI to allow users of Sora, the AI video platform, to generate videos featuring more than 200 characters from Disney franchises. Meanwhile, governments around the world rewrite copyright rules for content-intensive machines. Is training AI on copyrighted works fair use? As with any billion dollar legal matter, it depends. —Michelle Kim
14. GEO
Just a few years ago, an entire industry was built around helping websites rank high in search results (okay, 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 maximize their visibility in AI, whether in AI-enhanced search results like Google’s. AI Insights or in the LLM answers. It’s no wonder they’re panicked. We already know that news agencies have experienced a colossal fall In search-generated web trafficand AI companies are working on ways to cut out middlemen and allow their users to visit sites directly from their platforms. It’s time to adapt or die. —Rhiannon Williams