FT: Sources say OpenAI cutting corners on AI safety

OpenAI has drastically cut the amount of time allocated for safety testing of its AI models due to competitive pressure from rivals like Meta and Google, according to a report in the Financial Times.

OpenAI is due to release the large version of its o3 “reasoning” model soon, and sources close to the company are raising alarms that the model was not subjected to the same rigorous six-month-long safety assessment that earlier models like GPT-4 were subject to.

The company told the FT that the reduced testing time was a result of efficiencies and automation.

OpenAI slashes AI model safety testing time

The company told the FT that the reduced testing time was a result of efficiencies and automation.

More Tech

See all Tech

Jon Keegan12h

OpenAI reportedly poaching key Apple designers, using Apple manufacturing partners for AI gadgets

New details are emerging about the mysterious AI gadgets being designed by former Apple design chief Jony Ive since OpenAI purchased his startup “io” in May.

According to a report by The Information, Ive’s team has recruited several key Apple design and hardware employees to work on the gadgets. The Information reported some details of the devices:

“One of the products OpenAI has talked to suppliers about making resembles a smart speaker without a display, the people said. OpenAI has also considered building glasses, a digital voice recorder and a wearable pin, and is targeting late 2026 or early 2027 for the release of its first devices, one of the people said.”

OpenAI is also turning to Apple’s Chinese manufacturing partners to build the products, having signed contracts with Luxshare, and has been in talks with Goertek, per the report.

OpenAI Raids Apple for Hardware Talent, Manufacturing Partners

“One of the products OpenAI has talked to suppliers about making resembles a smart speaker without a display, the people said. OpenAI has also considered building glasses, a digital voice recorder and a wearable pin, and is targeting late 2026 or early 2027 for the release of its first devices, one of the people said.”

OpenAI is also turning to Apple’s Chinese manufacturing partners to build the products, having signed contracts with Luxshare, and has been in talks with Goertek, per the report.

Are Ray-Ban Meta glasses really a hit?

We checked how it stacks up to iconic gadgets, and the results are mixed.

Jon Keegan, Rani Molla12h

Rani Molla14h

Zuckerberg: AI might be a bubble but “misspending a couple of hundred billion” is worth it to achieve superintelligence

“It’s quite possible” that AI is a bubble, Meta CEO Mark Zuckerberg told tech journalist Alex Heath, formerly of The Verge, on his new podcast, “Access,” and for his newsletter, Sources. That isn’t stopping Zuckerberg’s social media company from going all in on AI in hopes of achieving superintelligence, aka AI that’s smarter than humans.

“If we end up misspending a couple of hundred billion dollars, I think that that is going to be very unfortunate, obviously,” said Zuckerberg, who’s shelling out $600 billion on US data centers and infrastructure through 2028. “But what I’d say is I actually think the risk is higher on the other side.”

“The risk, at least for a company like Meta, is probably in not being aggressive enough rather than being somewhat too aggressive,” he added.

I saw Meta's new AI lab

“The risk, at least for a company like Meta, is probably in not being aggressive enough rather than being somewhat too aggressive,” he added.

Rani Molla

15h

People are rushing to buy Apple’s iPhone 17, but it’s probably not because of the iPhone 17

More people went to Apple’s website when preorders for the new iPhone began than have since 2022.

Rani Molla8h

Grok has 64 million monthly users while ChatGPT has 700 million weekly users

Daddy, it seems, is very much not home.

CEO Elon Musk spent the majority of his time at xAI this summer rather than at Tesla, where he recently claimed to have shifted his focus, The New York Times reports. The piece is full of other great details on his AI startup — read it all — but here are some notable tidbits from the story and from one of its reporters, Kate Conger, who shared extras on social media:

xAI’s Grok has 64 million monthly users, compared with OpenAI’s ChatGPT, which has about 700 million weekly users. Musk is currently suing OpenAI and Apple over what he says is unfavorable positioning on the iOS App Store.
Musk wanted Grok to be less woke and more popular, a command that led it to post antisemitic remarks and call itself “MechaHitler.”
Musk plans on building a Microsoft competitor called “Macrohard,” something he said he’s painting on the roof of xAI’s new Memphis data center.
xAI’s execs said after Grok 4, the next model will be called Grok 420.

UPDATE (September 19): Corrected headline of piece to reflect ChatGPT has 700 million weekly users, not daily.

Elon Musk Has Focused on xAI Since Leaving Washington

xAI’s Grok has 64 million monthly users, compared with OpenAI’s ChatGPT, which has about 700 million weekly users. Musk is currently suing OpenAI and Apple over what he says is unfavorable positioning on the iOS App Store.
Musk wanted Grok to be less woke and more popular, a command that led it to post antisemitic remarks and call itself “MechaHitler.”
Musk plans on building a Microsoft competitor called “Macrohard,” something he said he’s painting on the roof of xAI’s new Memphis data center.
xAI’s execs said after Grok 4, the next model will be called Grok 420.

UPDATE (September 19): Corrected headline of piece to reflect ChatGPT has 700 million weekly users, not daily.

Latest Stories

Sherwood Media, LLC produces fresh and unique perspectives on topical financial news and is a fully owned subsidiary of Robinhood Markets, Inc., and any views expressed here do not necessarily reflect the views of any other Robinhood affiliate, including Robinhood Markets, Inc., Robinhood Financial LLC, Robinhood Securities, LLC, Robinhood Crypto, LLC, or Robinhood Money, LLC.