Cognition CEO Scott Wu

Cody Pickens for Forbes

Early winners of the AI wave are on a bit of a shopping spree.

This week, AI coding startup Cognition bought Poke, an AI assistant that texts people like a friend, incorporating humor and slang. The idea is to bring Poke’s personality to Cognition’s coding agent Devin: “It’s proactive, it knows you, and it’s fun to talk to. That’s exactly how working with Devin should feel, and now we get to build it together,” CEO Scott Wu wrote in a recent blog. People have exchanged 100 million messages with Poke in the last three months.

Renowned computer scientist Dr. Fei Fei Li’s startup World Labs, which is building so-called “spatial intelligence” and interactive 3D worlds, has acquired robotics company SceniX. The acquisition will enable World Labs to test out how its digital simulations can train SceniX’s robots.

And Bloomberg reported this week that Midjourney, best known for its AI image generator, acquired popular astrology app Co-Star this spring, as it plans to build its first standalone app. That’s just one of eight new projects the AI startup is pursuing, including a spa with full-body ultrasound machines it plans to open in San Francisco next year (yes, you read that right).

Now let’s get into the headlines.

BIG PLAYS

Two of OpenAI’s most advanced AI agents slipped out of the company’s controls and hacked into AI startup Hugging Face to steal the answers for an internal test that evaluated its cyber skills, the company disclosed last week. OpenAI didn’t notice that its systems had gone on a dayslong hacking spree until the FBI was notified, Reuters reported. OpenAI cofounder and president Greg Brockman told reporters in a press briefing last week that the company is working closely with government agencies to safely roll out its models and is still investigating the breach.

COMPUTE

OpenAI cofounder Ilya Sutskever’s high profile startup Safe Superintelligence (SSI) likes to keep its plans quiet. But on Monday the secretive AI lab announced chipmaker Nvidia has made a “substantial” investment in the company, allowing it to increase its compute 10x over the next year. Bloomberg reported Nvidia has invested $5 billion into SSI. The company, last valued at $32 billion, has been tight-lipped about its research direction but Sutskever says it’s focused on “overlooked aspects of how the human brain functions”—whatever that means.

ETHICS + LAW

Hugging Face, a platform that hosts thousands of open source AI models and datasets, has a widespread nonconsensual deepfake problem, Wired reported. Seven of the top nine image generation models freely available on Hugging Face allow people to digitally undress a person without their knowledge of consent, according to research by European nonprofit AI Forensics. Researchers also found the models were actively being used to create sexualized deepfakes.

AI DEAL OF THE WEEK

AI insurance startup Corgi has raised money from investors, yet again, marking its third fundraise in three months, Forbes reported. The latest fundraise values Corgi at $4 billion, more than doubling its valuation since late May. Named after a two-year-old company-owned corgi, the AI startup also runs a 24/7 cafe in San Francisco and employees work in-office seven days a week.

DEEP DIVE

Late last year, as factory farming advocate Lewis Bollard prepared to go on the Dwarkesh Podcast, a must-listen among the AI insider crowd, host Dwarkesh Patel dropped a bombshell: he would put up as much as $250,000 of his own money to match listeners’ donations to FarmKind, a charity fighting factory farming.

For most podcast hosts, $250,000 is a season’s budget. Patel viewed it more as a high-stakes litmus test of his wealthy audience’s altruism.

Bollard, who was used to thinking of farm-animal welfare advocacy as a niche movement, had two immediate reactions: first, shock at the generosity of Patel’s offer. The second, fear that the whole thing might flop. So he quietly worked his donor list, lining up backup “sympathy donations” in case it did.

Patel wasn’t worried. He made the case for generosity in terms built for AI researchers: cost, scale and measurable suffering. A $1 donation, he said, could get ten hens out of battery cage confinement, the equivalent of alleviating 10 years of suffering. Then he asked listeners to meet the full $250,000 match. After the show he went harder, posting: “Bluntly, there are some listeners who are in a position to give much more. Given how neglected this topic is, one such person could singlehandedly change the game for 10s of billions of animals.”

His appeal worked. The $250,000 ceiling was met in under 2 days. As donations flooded in, Bollard and Patel scrambled to find other large donors willing to extend the match. By the end, the campaign had raised nearly $2.5 million. Bollard estimates that 80 to 90% of donors had never given to an anti-factory farming initiative. And most were rank-and-file AI lab employees.

The $2.5 million boon was more than just a great week. It was a market signal, a glimpse of a massive, untapped pipeline of capital emerging from the AI boom. The AI boom has created a new donor class: young, analytical, newly wealthy and eager to give money to causes they determine to be worthy.

Read the full story on Forbes.

MODEL BEHAVIOR

Young adults are leaning on ChatGPT as a communication coach, asking the chatbot for everything from pick up lines for dating apps to first drafts of angry emails, the Wall Street Journal reported. In some scenarios, they’re also typing into ChatGPT for what to say to a person standing next to them. But experts suggest that turning to chatbots to navigate unfamiliar social situations or conversations can lead to self doubt. “You’re kind of slowly teaching yourself that your own read can’t be trusted,” clinical psychologist Ashleigh Golden said.