What's trending in AI right now, from the app charts to the community feeds. Real links, our take.
AI safety warnings went mainstream this week
Two things happened in quick succession. A researcher named Jacob Coxon left Anthropic after three years and posted publicly that the labs building this technology are "gambling with our lives", saying the people doing this work genuinely believe it could kill everyone by end of decade but are racing ahead anyway because they think competitors won't stop. Days later, Anthropic CEO Dario Amodei published an essay calling for the whole industry to slow down, warning that AI agent swarms could seize control of large parts of the internet within a year. OpenAI's Sam Altman agreed publicly. Videos about both stories are pulling tens of thousands of views per hour right now. If friends outside tech are suddenly asking you about AI risk, this is likely why.
Someone may be reading your ChatGPT conversations
404 Media published a detailed report on OpenAI's internal "Project Lily", which pays hundreds of outside contractors to read real user conversations and rate ChatGPT's replies. The goal is model improvement, which is common practice across the industry. What surprised people: OpenAI does not surface this clearly, and one contractor told 404 Media that users probably have no idea it happens. OpenAI says it strips identifying information before prompts reach reviewers, but acknowledged sensitive details can still get through. Reddit threads on this have been climbing all week, which tracks with how big the gap feels between "I tell this thing everything" and "a contractor somewhere is reading this."
AI study apps are climbing as school starts
Two education apps jumped noticeably in the charts this week. Gauth, ByteDance's homework helper, hit the number two Education spot after launching a feature that lets students generate personalized courses on any topic they choose, on top of its existing library of 200-plus AI math courses. Gizmo, a flashcard and quiz app, climbed four spots after announcing 13 million users and a $22 million funding round earlier this year. Back-to-school timing is part of it, but AI study tools occupying the top slots in Education charts is a real shift from even two years ago.
ChatGPT Plus users are hitting a wall
A Reddit thread titled "plus is becoming too unstable for actual work" has been getting consistent traction, and the complaints are grounded in something real. OpenAI has logged more than 60 incidents in the past 90 days, with paid subscribers hit hardest because the work-mode features they specifically pay for are the ones going offline. People who upgraded to Plus for reliability are now questioning whether the tier is delivering what it promised.
Self-hosted AI just got a real mobile app
Conduit, a native mobile client for Open WebUI and self-hosted AI models, made the biggest single chart jump of the week, climbing 12 spots in Productivity. It is built for people who run their own AI setup and want phone access without routing through a major cloud service. The app connects only to servers you control and ships with no ads and no analytics. A 12-spot climb suggests the audience for self-hosted AI is bigger and more willing to pay for a polished interface than it was even a year ago.