Ad
Skip to content

New Deepseek technique balances signal flow and learning capacity in large AI models

DeepSeek researchers have developed a technique that makes training large language models more stable. The approach uses mathematical constraints to solve a well-known problem with expanded network architectures.

China captured the global lead in open-weight AI development during 2025, Stanford analysis shows

Chinese open-weight AI is conquering the world: According to a Stanford analysis, models from China have already overtaken their US counterparts in distribution and adoption. But with success come growing geopolitical and security risks.

AI benchmarks are broken and the industry keeps using them anyway, study finds

Benchmarks are supposed to measure AI model performance objectively. But according to an analysis by Epoch AI, results depend heavily on how the test is run. The research organization identifies numerous variables that are rarely disclosed but significantly affect outcomes.

Read full article about: Global AI compute hits 15 million H100 equivalents, Epoch AI finds

Epoch AI has released a comprehensive database of AI chip sales showing that global computing capacity now exceeds 15 million H100 equivalents. This metric compares the performance of various chips to Nvidia's H100 processor. The data, published on January 8, 2026, reveals that Nvidia's new B300 chip now generates the majority of the company's AI revenue, while the older H100 has dropped below ten percent. The analysis covers chips from Nvidia, Google, Amazon, AMD, and Huawei.

Epoch AI estimates this hardware collectively requires over 10 gigawatts of power - roughly twice what New York City consumes. The figures are based on financial reports and analyst estimates, since exact sales numbers are often not disclosed directly. The dataset is freely available and aims to bring transparency to computing capacity and energy consumption.

Read full article about: Musk's lawsuit against OpenAI heads to trial

Elon Musk's lawsuit against OpenAI and CEO Sam Altman is going to trial. A California federal judge announced Wednesday that she intends to reject attempts by Altman's lawyers to dismiss the case. Judge Yvonne Gonzalez Rogers said during the hearing in Oakland that there is ample evidence to proceed.

Musk accuses OpenAI of deceiving him about its shift from a nonprofit to a for-profit structure. He says he donated $38 million to the company. The trial is scheduled for March. OpenAI denies the allegations, calling the lawsuit baseless and part of ongoing harassment by Musk.

The company claims Musk was informed about its profit plans back in 2018. Musk co-founded OpenAI in 2015 and left the company in 2018.

Read full article about: Minimax stock doubles on Hong Kong debut

Shares of Chinese AI startup Minimax doubled in value during their Hong Kong Stock Exchange debut. The stock closed up 109 percent at 345 Hong Kong dollars, CNBC reports. Minimax significantly outperformed local rival Zhipu AI, whose shares gained just 13 percent the day before. The IPO raised around $620 million for Minimax.

The company, backed by Alibaba and Tencent, develops language models for chatbots and video generation. Despite having over 200 million users and revenue jumping to $53.4 million, Minimax reported a $512 million loss for the first nine months of 2025. The company says it's funneling earnings into research. Meanwhile, Disney, Universal, and Warner Bros have been suing Minimax for copyright infringement since September 2025.

Comment Source: CNBC

Researchers extract up to 96% of Harry Potter word-for-word from leading AI models

Harry Potter, Game of Thrones, 1984: researchers pulled nearly complete books from commercial language models. Two of the four systems tested didn’t even put up a fight. The findings could shape ongoing copyright lawsuits against AI companies.