This week in AI: Musk offers for Openai

Welcome to the regular AI bulletin of Hiya, People, TechCrunch. If you like this in the Inbox every Wednesday, please register here.

Billionaires fight again.

Monday, the richest man in the world Elon Musk, Offered to receive non-profit This manages Openai to $ 97.4 billion effectively. In response to the azlin proposal, Openai CEO Sam Altman is the author of Monday before A burnt writing in XWriting, “Thank you, but we will get twitter for $ 9.74 billion if you want.” (Musk and investors are celebriously Twitter purchased for $ 44 billion In 2022.)

The proposal, serious or absence of musk may complicate the efforts of an openness to become a corporation for profit for profit in two years. Now the Board of Openai will have to demonstrate a discount (such as althania), non-profit assets, including non-profit assets, including non-profit assets, including non-profit assets, including non-profit assets, including Openai’s research.

Openai, Musk’s proposal said that this Musk and Altman’s hostile attempt to seize the enemy Not the best of friends. You can also claim that the proposal of the musk is not valid, because Openai is already in the process of reconstruction. Or can Openai Accompany whether or not funds.

One Statement TuesdayAndy Nussbaum, the foreign advice representing the Board of Openai, Musk’s offer “(Openai’s)” not for nonprofit “for sale,” Nussbaum added: “Sincerely, Openai’s mission It’s up to an opponent to decide what in the best interests. ”

My colleague Maxwell Zeff and I He wrote a piece of more detail What to expect in the coming weeks. But guaranteed, Muskin offer – not to mention it The claim that is against Openai is false behavior – The fierce courtroom promises to bend the neck.

News

Apple’s new robot: Apple created a research robot that took a page from Pixar’s PlayBook. The company’s robot lamp operates as a more kinetic version of a homepod or other clever dynamic. The person facing the lamp asked for a request and the robot responds to the SIR.

AI makes us dumb?: Researchers recently published a research despite the use of a generative AI. When we trust in AI, too much to think for us, we get worse in solving problems when AI fails.

AI, perhaps: In a new essay, Altman said that the benefits of the EU’s benefits should not be widespread, and as a “strange budget” such as “strange budget”, said that “weirdoic budget” has made it possible to use “everything on earth” AI. “

Christie’s dispute: The subtle art auction house sold an art created by the EU at the beginning of Kristie. However, the first show soon refer to the work created with the AI, an announcement for the announcement and auctions that are met with mixed feedback.

Better than gold: Google Deepmind, the AI ​​system developed by Google’s leading AI research laboratory, the average gold medalist in the international math competition in solving geometry problems.

Research paper of the week

We know that most of AI models can not be reliably fulfilling the main tasks such as school-level math problems. What we don’t always know cause behind the failures. According to the researcher team in Mit Clael, the wrong backlers may be partially guilty.

In a new study, when MIT Clai researchers make real mistakes of more than 50% of today’s highest performance models, more than 50% of “model mistakes” mistakenly created uncertain questions in these trends they saw.

“If we want to measure the validity of the model properly, we need to re-think how we set up patches to minimize label errors,” he said One of the researchers, MIT teacher staff and Openai Staffer Alexander Madry, In an article in X. “It’s just the first step.”

The model of the week

You’ve heard about deep deposit before. What about the boredom of daily scenes? This idea is that Bored Reality Hunyuan Lora (Boreal-HL)A beautiful adjustable AI video generator superior to creating videos … well, pretty banal items.

Boreal-HL, people who eat ice cream, people who are maneuvering people, people in dining congregations, people in weddings, other secular slices, other secular slices can create managers and executives. This correspondent finds the absurdity of the merry thing – especially the more impossible to run. Boreal-HL for at least five minutes to create a single clip.

Catch a bag

Thanks to the latest improvements in the EI efficiency, it becomes better – and it is easier – train highly developed models.

In a new paperShanghai Jiao Tong University, researchers and SII company demonstrates 100,000 more information that a program that has been a training on 817 “low-end training samples.” The team claims that the model can answer certain questions that he did not see during the training process shows the “outside the domain”.

The study follows in a heels Stanford LED Project It was possible to create a model of Openai, Openai O1 “Justification” of O1 “Justification”.