In a newly published paper, Chinese AI firm DeepSeek has detailed a new approach to training AI models it calls “Manifold-Constrained Hyper-Connections.” The method allows training larger and larger AIs while reducing needed computer time and, therefore, energy consumption.
Tag: DeepSeek
DOT COM BUST 2.0
On 28 January 2025, The Trends Journal alerted subscribers about this little-known Chinese AI startup, DeepSeek, that sent a bolt across the AI-driven stock market’s bow when it announced that it trained a free AI assistant for $6 million instead of the $100-million-plus that Open AI and other Western tech companies had to spend on their products.
CHINA STEPS UP CRACKDOWN ON U.S.-MADE COMPUTER CHIPS
After pressuring Chinese tech companies to use only domestically made computer chips, Beijing in mid-September ordered a halt to all purchases of chips made in the U.S.
DEEPSEEK REVEALS ANOTHER AI THAT’S CHEAPER TO OPERATE
Chinese research firm DeepSeek has released a new AI model that it says can be operated more cheaply than the large language models common in the West.
DEEPSEEK SAYS ITS FLAGSHIP AI COST ONLY $294,000 TO TRAIN
DeepSeek, the Chinese AI developer that jolted the industry in January by releasing a world-class model that it said it built on the cheap, now says it spent just $294,000 to train it.
DEEPSEEK IS UNABLE TO TRAIN ITS NEXT MODEL ON CHINESE CHIPS
DeepSeek, the Chinese AI developer that stunned the world in January with its world-class large language model made on the cheap, has been forced to delay the release of its R2 reasoning AI.
CHINA IS GAINING ON THE U.S. IN THE GLOBAL AI RACE
A growing number of businesses and universities in Africa, Asia, Europe, and the Middle East are adopting DeepSeek and other AI systems from Chinese developers and turning away from U.S. models, The Wall Street Journal reported.
CHINESE STARTUP SAYS IT’S BETTER THAN DEEPSEEK
MiniMax, yet another Chinese AI developer, says its new, open-source MiniMax-M1 model, is more efficient than closed-source Chinese competitors and also bests DeepSeek's latest R1-0528 model in several standard tests.
LEADING AI MODELS “COLLAPSE” UNDER HARD PROBLEMS, STUDY SAYS
When struggling with highly complex problems, “reasoning” AIs from DeepSeek and OpenAI suffer “complete accuracy collapse,” according to a new study released by Apple.
DEEPSEEK RECASTS CHINA’S AI INDUSTRY
Since announcing its open-source, world-class, cheap AI in January, DeepSeek has seen its technology restructure China’s AI industry.









