This Week in AI: Musk bids for OpenAI

Hia, people, welcome to TechCrunch regular AI newsletter. If you want it in your inbox every Wednesday, sign up HereThe

Billioners are fighting again.

Monday, the richest person in the world, Elon Kasturi, Offer to buy non -profit It effectively operates the opeina for .4 97.4 billion dollars. In response to Kastur’s proposal, OpenAI’s CEO Sam Altman composed before Monday A slick post at x“Don’t thank you, but if you want to buy Twitter for $ 9.74 billion.” (Kasturi and investors famously Bought Twitter for $ 44 billion In 2022)

The bid of the mask, serious or not, can complicate the attempt to convert the profitable public benefits to the corporation within two years of the opening. Now the OpenAI Board will have to show that the OpenAI’s non -profit (eg, Ultman) is not underprofit for the exemption of non -profit resources, including IPs from OpenAI research.

Opena can make the case that Kasturi bid is an adverse techover attempt that Kasturi and Ultman Friends are not the bestThe It can also argue that the offer of musk is not credible because the openness is already within a reconstruction process. Or Opena could Challenge Kasturi on whether he has fundsThe

A The statement is TuesdayOPENAIA Board Representatives Andy Nusboum said that Kasturi’s bid “does not set any prices for [OpenAI’s] Non -profit “and not for non -profit” for sale “. Nusboum also added,” With reverence, it does not depend on any competitor to decide what is in the best interest of the mission. ”

My colleague Maxwell Jeff and I Wrote more detailed pieces What to expect over the next weeks. However guaranteed, musk offers – not to mention it What the ongoing case against the opeina claims is the fraudulent behavior – Promise to quarrel the fatal courtroom.

News

Apple’s new robot: Apple has created a research robot that takes a page from Pixar’s PlayBook. The company’s robotic lamp acts as a more motion of homepod or other smart speakers. The person in the face of Pradeep asks a question, and the robot responds to Siri’s voice.

AI is doing us dumb ?: Researchers recently revealed a research on how the use of generator AI in the workplace affects critical thinking skills. It has shown that when we rely too much on AI to think for us, when AI fails we get worse in solving the problem ourselves.

AI for everyone, perhaps: In a new article on his personal blog, Ultman acknowledges that AI’s benefits cannot be widely distributed-and said that Open is open to “strange-sounding” concepts like a “count budget” to enable everyone on earth to use a lot of people on earth. “

Christie’s debate: Fine Art Auction House Christie has previously sold AI-exposed art. However, soon it was planned to be dedicated to the work done by AI soon, a declaration that was filled with mixed reviews – and an application called for the cancellation of the auction.

Better than gold: Google’s leading AI research lab, an AI system built by Google Dipmind, seems to have surpassed the average gold medal to solve the geometry problem in an international mathematics competition.

Weekly

We know that most AI models cannot perform the initial tasks reliably to solve grade-school-level math problems. We don’t always know which Cause Behind their failure. According to a group of researchers at MIT CSAL, false benchmarks can be part of blaming.

In a new study, MIT Sicel researchers have discovered that today’s leading performance models are still true at the popular AI criteria, more than 50% of the “model error” is actually due to misleading and vague questions in these benches.

“If we want to determine the appropriate amount of the model reliability we need to rewrite on how to make benchmarks to reduce label errors,” said A researcher, a member of the Faculty of MIT and Alexander Madri, OpenAI worker, In a post at XThe “This is simply the first step.”

Week

You’ve heard about Dipfkes before. But what about the depth of the daily scene of the boring? That’s the idea behind Annoying reality Hunuan Lora (Borial-HL)A subtle AI video generator that is… the good, beautiful banner stuff out of the video.

Boreal-HL ice cream can make tourists clip, meat barbeng, people at lunch meetings, conference speeches, couples at the wedding and other worldly pieces of life in life This reporter finds the irony of the smiling thing in pieces – considering how irrational it is to run it especially. It takes at least five minutes to make a single clip to create a single clip.

To grab the bag

Thanks to the latest epoch of AI skills, it is cheaper for training highly sophisticated models – and is becoming easier.

On a new paperResearchers at Shanghai Ziao Tong University and an AI company called SII prove that only 817 “curate training samples” trained models “can exceed 100x more data trained models. The team claimed that their models were able to answer certain questions even during the training process, which they showed as “outside the domain”.

The study follows a heel Stanford It has shown that the OpenAI and 1 “logic” model is possible to create a “Open” model by competing below $ 50.