In-depth articles with more detailed investigations, interesting thought pieces, and more.
OpenAI has released its first detailed study on ChatGPT usage, revealing that most people use the chatbot for personal matters rather than work. The research shows that “practical guidance,” such as asking for how-to advice or help with schoolwork, is the most common reason people turn to the AI. Gerrit De Vynck reports for the …
The two leading AI assistants, ChatGPT and Claude, have been built with fundamentally opposite philosophies on how to remember user interactions. ChatGPT’s memory is designed to be automatic and invisible, creating a personalized experience with zero effort from the user. Claude’s initial approach, however, gave users explicit control over when to use its memory function. …
Thousands of human workers who train and moderate Google’s AI models like Gemini face grueling deadlines, low pay, and exposure to distressing content. These “raters,” hired through contractors like GlobalLogic, are essential for making chatbots seem intelligent and safe but often feel invisible and expendable. This is reported by The Guardian after speaking with ten …
Large language models like ChatGPT sometimes generate false information (“hallucinations”) because their evaluation systems reward guessing over admitting uncertainty. In an official post, the company OpenAI reports that this incentive structure is a fundamental challenge for all current AI models. Hallucinations can occur even with seemingly simple questions. For example, a chatbot gave three different …
Tech companies have downloaded more than 15 million videos from over two million YouTube channels to train their artificial intelligence models. Alex Reisner reports for The Atlantic that this practice often occurs without the creators’ permission and violates the platform’s terms of service. The investigation identified at least 13 data sets containing the scraped content, …
A coalition of major web publishers, including Reddit, Yahoo, and Quora, has launched a new system called Really Simple Licensing (RSL). The standard aims to make AI companies pay for using online content to train their artificial intelligence models. The RSL Standard builds on the existing `robots.txt` protocol, a file that gives instructions to web …
Videos created using artificial intelligence are flooding social media platforms and generating substantial revenue for their creators, despite widespread criticism of the low-quality content. John Ruwitch reports for NPR. Mark Lawrence I Garilao, a 21-year-old computer science student from the Philippines, earns up to $9,000 per month creating simple animated videos featuring a kitten character. …
A widely misunderstood statistic from a new MIT report suggests corporate artificial intelligence projects are failing, but the study reveals the opposite. Michael Nuñez reports for VentureBeat that a “shadow AI economy” is thriving as employees successfully use personal AI tools for their work, outpacing official corporate initiatives. The study from MIT’s Project NANDA found …
Colleen Jones reports for Content Science Review that despite the hype around generative AI, most leaders struggle to achieve measurable business benefits. The primary reason is that AI, unlike previous technologies, is uniquely dependent on content for both its input and output. Jones identifies seven key barriers that prevent companies from succeeding with AI. A …
AI web crawlers are overwhelming websites and disrupting the economic foundation of online publishing, according to Cloudflare CEO Matthew Prince. The cybersecurity company executive has launched a “pay-per-crawl” initiative to force AI companies to compensate content creators whose sites they harvest for training data. Prince argues that AI chatbots have fundamentally broken the traditional web …