Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Hello, people, welcome to the regular newsletter of Techcrunch. If you want this in your inbox every Wednesday, register here.
Billionaires are fighting again.
On Monday, Elon Musk, the richest man in the world, I offered the purchase of non -profit organizations This effectively rules Openai for $ 97.4 billion. In response to the Musk offer, the CEO of OpenAi Sam Al -Tamman earlier on Monday Sficiq’s post on xWriting, “No thanks to you, but we will buy Twitter for $ 9.74 billion if you want.” (Musk and investors are famous Buy Twitter for $ 44 billion In 2022.)
MUSK, serious or not, may complicate Openai’s efforts to convert to the Public Paper benefits company within two years. Now the OpenAI board will have to prove that it is not unbearable from Openai by delivering non -profit assets, including IP from Openai’s Research, to the inside (for example, Altman) to get a discount.
Openai can show that the Musk’s offer is an attempt to hostile acquisition as Musk and Altman Not the best friends. It can also argue that Musk’s width is not credible because Openai is already in the restructuring process. Or openai The challenge of musk on whether he has the money.
in Tuesday statement“The MUSK offer does not specify the value of the non -profit organizations (Openai’s) and that non -profit organizations” are not for sale. ” Openai mission. ”
My colleague Maxwell is false and me He wrote a more detailed piece On what can be expected in the coming weeks. But guaranteed, the musk show – let alone The ongoing lawsuit against Openai on what it claims is a fraudulent behavior – Promises to make a fierce courtroom fight.
news
New Apple robot: Apple created a robot for the search that takes a page of Playbook’s Pixar. The company’s automatic lamp works as a more movement version of HomePOD or other smart speakers. The person who faces the lamp asks, and the robot responds in a Siri voice.
Does Amnesty International make us stupid?: Researchers recently published a study looking at how artificial intelligence affects working on critical thinking skills. It was found that when we rely a lot on artificial intelligence to think for us, we are getting worse to solve problems ourselves when artificial intelligence fails.
Amnesty International for All, perhaps: In a new article on his personal blog, Altman admitted that the benefits of artificial intelligence may not be widely distributed-he said that Openai is open to “strange sounding” ideas such as “account budget” to enable everyone on Earth to use many Amnesty International.
Christie’s controversy: The house of Fine Arts, Christie, has sold the art created by artificial intelligence before. But soon you plan to hold her first presentation only for the works created with artificial intelligence, a declaration that was met with mixed – and broad reviews calling for the abolition of the auction.
Best gold: It seems that the AI system developed by Google DeepMind, a Gogle’s leading AI research laboratory, has exceeded the average gold medal in solving engineering problems in an international mathematical competition.
Search paper in the week
We know that most artificial intelligence models cannot reliably perform basic tasks, such as solving mathematics problems at the level of school. What we don’t always know is the reason Behind their failures. According to a team of researchers at MIT CSAIL, the wrong criteria may be part of it.
In a new study, MIT CSAIL researchers found that although the higher models today are still making real errors on common artificial intelligence standards, more than 50 % of “typical errors” are actually caused by wrong and mysterious questions in those standards.
“If we want to properly determine the reliability of the model, we need to rethink how to build standards to reduce stickers’ errors,” he said. One of the researchers, faculty members at the Massachusetts Institute of Technology and Openai Aleksander Madry, In a post on X. “This is just a first step.”
Week model
I heard about Deepfakes before. But what about Deepfakes from boring daily scenes? This is the idea behind Hunyuan Lora (Boreal-HL)A video of artificial intelligence that exceeds the creation of videos from … well, beautiful ordinary things.
Boreal-HL can generate clips from tourists who take ice cream, people who make barbecue parties, people in lunch meetings, executives who hold letters in conferences, weddings at weddings, and other worldly life segments. This correspondent finds the absurdity of the fun thing – especially given the extent of my lack of work. Boreal-HL matter requires at least five minutes to create one section.
Grab a bag
Thanks to the recent breakthroughs in the efficiency of artificial intelligence, it has become cheaper – and easier – to train very advanced models.
In a new paperResearchers at the University of Shanghai Jiao Tong and Amnesty International, which is called SII, explains that the model trained on only 817 of the “coordinated training sample” can outperform the trained models over 100x other data. The team claims that their model was able to answer some of the questions that he did not see during the training process, while showing what they call the capabilities “outside the field”.
Follow the study on the wake of a A project led by Stanford This was found that it is possible to create an Openai “Openai” model from Openai for less than $ 50.