I’m an Atlantic staff writer and tech investigator working to improve our public discussion about the statistical models known as AI.
There's a worldwide shortage of computer memory that's driving up the price of computers, and new data centers are springing up everywhere. It's all happening because generative AI is a poorly engineered technology. In short, it does not scale.
AI developers are using millions of pirated songs to train products that can generate songs that are clear rip-offs of songs they were trained on. Article includes a dataset search tool and samples of similar songs.
AI companies claim their chatbots can reason about problems, but the industry is just exploiting a trick that was invented by gamers on 4chan in 2020.
Tech companies believe in intellectual property, but not yours.
AI doesn’t “learn” like a human, it makes copies of its training data, and uses those copies in its outputs. My investigation presents crucial research that the AI industry has been trying to keep hidden.
A little-known non-profit has been scraping millions of paywalled articles, lying to publishers about it, and taking donations from AI companies.
AI companies including Microsoft, Meta, Amazon, Nvidia, Runway, ByteDance, Snap, and Tencent have downloaded millions of videos from YouTube to train video-generating products that compete with YouTubers. Article includes a dataset search tool.
YouTube is adding subtle visual effects to make normal videos appear AI-generated. Why?
Two recent rulings in copyright infringement lawsuits make conflicting statements about how the technology works and what it means for creators.
Chatbots are increasingly used to control what information we can access. In February 2025, xAI employees instructed the company’s chatbot to ignore web pages that claim Elon Musk or Donald Trump spread misinformation.
Meta employees considered licensing books to train their LLMs but decided it would take too long. Mark Zuckerberg gave the green light to steal them instead. Article includes a dataset search tool.
Dialogue from movies and TV shows has been used by companies such as Apple and Anthropic to train AI systems. Article includes a dataset search tool.
This first-of-its-kind investigation reveals that pirated books are being used to train LLMs.