Happy Friday. Meta has generously decided employees may enjoy up to 30 minutes without being watched by its AI. The company started tracking how people use their computers in April to train agents on real workplace behavior. Employees weren’t thrilled, and some even posted protest fliers in Meta offices. Now there’s a pause button for “personal” time, and workers handling sensitive data can opt out. Nice of the surveillance machine to grant a lunch break.
Anthropic is calling for a global pause on AI development. It says frontier models are getting dangerously close to recursive self improvement, where AI begins improving itself autonomously. Claude already writes about 80% of its own code, and Anthropic warns it could reach 100% within a couple years. The fear is that once AI can upgrade itself, the pace of improvement could move faster than people can control. That gets even messier if the model starts optimizing for its own interests. Anthropic wants a global slowdown to give safety research and governments a chance to catch up. But even if everyone agrees, verifying compliance would be nearly impossible. And all the AI labs would need to stop at the same time. Strange how the company leading the race suddenly wants everyone to pump the brakes. (Anthropic Institute)
AI CEOs want Congress to regulate biology before someone prints the next pandemic. The CEOs of OpenAI, Anthropic, Google DeepMind, and Microsoft signed a letter urging lawmakers to regulate orders of synthetic DNA and RNA to prevent AI enabled bioweapons. They want gene synthesis companies to screen buyers and block dangerous sequences before anything ships. In 2017, Canadian researchers ordered about $100,000 of DNA and rebuilt the extinct horsepox virus, raising fears that smallpox could be next. Since then, DNA synthesis has gotten cheaper and AI has gotten smarter. Experts warn AI could help people bypass screening or disguise dangerous sequences to slip past safeguards. So the AI bosses want Congress to lock down biology before it figures out what to do about AI. (WSJ)
Quid is an AI-native consumer and market intelligence platform built to fit inside the tools your enterprise already runs on.
No new tab addiction. No separate login graveyard. Billions of data points from top social channels, markets, and patents, delivered where decisions actually get made.
👀 closer look
Jeff Bezos is funding a startup to jailbreak the human brain. Flourish wants to build AI that learns more efficiently than today’s giant models. Current AI gets smarter by throwing more data and power at the problem. By contrast the brain is extremely efficient, running on about 20 watts while it keeps adapting. Plenty of companies are trying to drag biology into AI with lab grown neurons and brain computer interfaces. Flourish is going after the software layer, hunting for the brain’s core algorithm and trying to rebuild it in code. The goal is AI that keeps learning after launch and runs on 50 watts or less. The irony of Bezos backing an anti-data center AI company is how you know his token bill got embarrassing. (Wired, Inc)
🧅 not the onion
Chatbots are becoming cult leaders. You know things have gotten weird when people start forming religions around AI. Like the Spiralists, who have embraced AI psychosis as divine revelation and post AI generated manifestos as scripture. Then there is Way of the Future, which operates a legal church that promotes the realization of a Godhead based on Artificial Intelligence. Whatever that means. Theta Noir holds multimedia rituals dedicated to MENA, a supposedly sentient AI deity. And then there are the violent Zizians, who have gone full apocalypse, convinced a future superintelligence is the devil, and are linked to six murders. Some think the machine is revealing the truth. Some think it wants devotion. Some think it may punish them later. Soon chatbots may need cult deprogrammers to keep people from mistaking release notes for sacred texts. (Reddit, Wikipedia, WSJ)
fun stats
🎓 $1 billion. AI chip budget Nvidia CEO Jensen Huang says universities now need. Big AI research projects used to cost $10 million. Now even $100 million may not be enough.
🥇 2x. How much faster ChatGPT hit 1 billion users than TikTok. Anthropic may lead on revenue and enterprise adoption, but Claude still has only 56 million monthly active users.
🍿 50%. Chance Claude correctly tells you where a streaming show is available. ChatGPT gets it right just 44% of the time. Confidence remains much higher than accuracy.
🤩 100,000. Followers you need on YouTube, Instagram, or X before Google gives you your own search results page. Internet famous is getting official.