A 'Green' AI App Is Going Viral on a Debunked Myth. Here's the Catch.

Pioneer, the summit where CX leaders redefine what’s possible, is on October 7th.

Join leaders from Fin, Anthropic, Clay, and Kalshi for an insightful conversation on the state of AI transformation.

You’ll discover how some of the most innovative minds in CX have transformed their organizations, learn how they think about CX, and hear how they're planning for what's next.

Join the conversation in San Francisco, or tune in virtually.

Reserve your spot

1.  A 'green' AI app is climbing the charts on a claim that does not hold up. An app called EcoGPT is spreading fast on Instagram with videos warning that AI data centers will drain the planet's drinking water, and it has shot up the App Store rankings in several countries as a result. The premise is an exaggeration. Sam Altman recently claimed that 38,000 ChatGPT queries use about as much water as growing one almond, and while fact-checkers note that data center water use is real and worth disclosing, nobody credible thinks it will empty the taps. Skeptics are now calling the app a scam that sells guilt rather than a better model.

2.  ChatGPT Work can now learn to write like you. OpenAI rolled out a Writing Style feature that reads your existing emails, messages, and documents in connected apps like Gmail, Slack, Google Drive, and SharePoint, then carries your phrasing, sign-offs, and even your capitalization habits into new drafts. It is live on the web for paid plans with access to ChatGPT Work. One open question is whether it weights a carefully written client email the same as a sloppy Slack message, so review the first few drafts closely.

3.  An AI-designed drug shows an early sign of slowing ageing in humans. Insilico Medicine published a study in Nature Biotechnology showing that its lung disease drug rentosertib lowered predicted biological age across six independent ageing clocks in 42 trial patients over 12 weeks. It is a secondary analysis of a small Phase 2a trial, and the authors say plainly that they cannot yet separate slower ageing from a lung that is simply getting better. The signal is interesting, but ageing clocks remain contested and the drug faces years of trials before any of this reaches a pharmacy.

The demos. OpenAI's GPT-6 Astra has been public for less than a week and the internet is full of people finding its ceiling. Matt Shumer asked it to build an Unreal Engine world populated by Astra-powered agents, then heard them talking to each other from the next room a day later (11.6M views). Others have fed it a mobile game ad and received a playable browser clone within half an hour.

The catch. Reporting from The Information, summarized by TechCrunch, ties Astra to a training method called recurrent depth, or opaque recurrence. Instead of writing out every reasoning step as readable text, the model cycles a query through the same internal layers several times and does part of the thinking in raw numbers. That is efficient. It also means a slice of the model's reasoning leaves no trace anyone can read.

The alarm. Ryan Greenblatt, chief scientist at Redwood Research, called it possibly the single worst development for AI safety to date, because investigators leaned heavily on readable chains of thought when they picked apart the OpenAI agent incident at Hugging Face this summer. His bigger fear is what comes next: scale the opaque part up and the model eventually reasons entirely in a space nobody can audit.

OpenAI's answer. Chief scientist Jakub Pachocki pushed back, saying Astra's computation depth sits within a factor of two of GPT-4 and that chain-of-thought monitoring remains a core research goal. But OpenAI's own safety testing reportedly found Astra harder to monitor than its predecessor, even as unwanted behavior on internal benchmarks dropped sharply. Better behaved, less legible. Both things are true at once.

The real question. Every lab wants the capability gains, and follow-up reporting says Anthropic and Google DeepMind are already discussing the same technique. The trade-off Astra exposed, power versus readability, is now an industry-wide choice rather than one company's experiment. Watch whether the labs agree on a limit before the market decides for them.

Not a robot. Astra cleared all 48 levels of "I'm Not a Robot," a game built entirely from the visual puzzles meant to prove you are human (2.1M views).

Brain explorer. Axultan, Head of Product at Higgsfield, used Astra to build a navigable 3D map of a fruit fly nervous system with more than 166,000 individual neurons you can zoom into and inspect. It is built on Google and HHMI Janelia's published research.

Back to Paint. Instead of asking Astra to generate a portrait, one user told it to open Paint in the browser and draw them by hand. The result is surprisingly good (6.5M views).

Gemini edits video. A creator broke down 10 Gemini prompts for turning raw footage into finished social clips, including camera-angle changes and scene recreations, no editing software required.

Calendar art. Someone asked GPT-6 to draw Michael Jackson using nothing but Google Calendar event blocks. It worked (1.1M views).

Build a floating sky kingdom in seconds

This Midjourney prompt produces a photorealistic celestial city above the clouds: pearl-white palaces, translucent bridges, and mountain-sized cloud banks lit from within. The negative cues at the end are what keep it from sliding into video game concept art. Try it in vertical format for a phone wallpaper or a story background.

Swap "sapphire blue and soft lavender" for "warm amber and rose gold", "deep emerald and pale mint", or "silver and midnight indigo" to change the mood entirely. For a moodier version, add "distant lightning inside the cloud banks" just before the lighting cues.