This week in AI: Musk Bids for Openai
Hiya, Friends, Welcome to the regular newsletter of Techcrunch AI. If you want this in your reception box every Wednesday, register here.
The billionaires are fighting again.
Monday, Elon Musk, the richest man in the world, offered to buy the non -profit organization This governs Openai effectively for $ 97.4 billion. In response to Musk’s offer, the CEO of Openai, Sam Altman, earlier on Monday, is written A cheeky post on xWriting: “No thanks, but we will buy Twitter for $ 9.74 billion if you wish.” (Musk and famous investors Bought Twitter for $ 44 billion in 2022.)
Musk’s offer, serious or not, can complicate OpenAi’s efforts to convert into a public public benefits company within two years. Now, the board of directors of Openai will have to demonstrate that it does not undergo an OpenAi non-profit by putting the assets of the non-profit organization, including the IP of the search for OPENAI, to an initiate (for example, Altman) for a discount.
Openai could argue that Musk’s offer is an attempt to take control of hostile since Musk and Altman are not the best friends. He could also say that Musk’s offer is not credible because Openai is already in the middle of a restructuring process. Where Openai could challenge musk on the question of whether he has the funds.
In a Tuesday declarationAndy Nussbaum, external lawyer representing the Board of Directors of OpenAI, said that Musk’s offer “did not set any value for [OpenAI’s] non -profit “and that the non -profit organization is” not for sale “. Nussbaum added: “Respectfully, it is not a competitor to decide what is in the best interest of the mission of Openai.”
My colleague Maxwell Zeff and I wrote a more detailed part What to expect in the coming weeks. But guaranteed, Musk’s offer – not to mention his A current trial against Openai on what he claims to be fraudulent conduct – Promises to make fierce audience room fights.
News
The new Apple robot: Apple has created a search robot that takes a page of the Pixar game book. The company’s robotic lamp works as a more kinetic version of a homepod or another intelligent speaker. The person facing the lamp requires a request, and the robot responds in Siri’s voice.
Does AI make us stupid?: Researchers recently published a study on how the use of work generating work affects critical thinking skills. He found that when we count too much on AI to think for us, we worse to solve the problems ourselves when AI fails.
For everyone, maybe: In a new test on his personal blog, Altman admitted that the advantages of AI may not be widely distributed – and said that Optai was open to “strange sounding” ideas as a “calculation budget” for “Allow everyone on Earth to use a lot of AI. “”
Christie’s controversy: The auction house of Fine-enchères Christie’s has already sold art generated by AI. But soon, he plans to hold his first show dedicated only to the works created with the AI, an announcement which met mixed criticism – and a petition calling for the cancellation of the auction.
Better than gold: An AI system developed by Google Deepmind, the main research laboratory on Google AI, seems to have exceeded the medium gold medalist in solving geometry problems in an international mathematics competition.
Week search document
We know that most AI models cannot do basic tasks reliably, such as solving mathematical problems at the level of the school. What we don’t always know is the reason Behind their failures. According to a team of MIT CSAIL researchers, erroneous benchmarks may be partly to blame.
In a new study, researchers from MIT CSAIL have found that even if today’s models of today always make real mistakes on popular AI landmarks, more than 50% of “model errors” are in Made caused by erroneous and ambiguous questions in these references.
“If we want to correctly quantify the reliability of the model, we must rethink the way we build benchmarks to minimize label errors,” said One of the researchers, a member of the faculty of the MIT and member of the Openai staff, Aleksander Madry, In a post on x. “This is only a first step.”
Model of the week
You have already heard of Deepfakes. But what about the depths of boring everyday scenes? This is the idea behind Boring reality HUNYUAN LORA (BOREAL-HL)A refined AI video generator that excels in creating videos of … well, fairly banal things.
Boreal-HL can generate tourist clips eating ice cream, meat barbecue people, people in lunch meetings, executives pronounced speeches during conferences, couples during weddings and other slices of life banal. This journalist finds the absurdity of the hilarious thing – in particular given the way it is impractical to run. You need Boreal-HL at least five minutes to generate a single clip.
Handbag
Thanks to the recent breakthroughs in the effectiveness of the AI, it becomes cheaper – and easier – to form very sophisticated models.
In a new newspaperResearchers at the University of Shanghai Jiao Tong and an AI company called SII demonstrate that a model formed on 817 “organized training samples” can surpass the models formed over 100 times more data. The team says that their model has even been able to answer certain questions that she had not seen during the training process, showing what they call the capacities “outside the field”.
The study follows the heels of a Project led by Stanford This revealed that it is possible to create an “open” model competing with the O1 “reasoning” model of Openai for less than $ 50.
