Latest Search
Quote
| Back Zoom + Zoom - | ||||
|
US Scholar: US-CN AI Race To Be a Tie; Costs Plunge as Other Countries Catch Up
Recommend 3 Positive 4 Negative 6 |
||||
|
||||
|
DeepSeek's release of an AI model has dealt a blow to the U.S. tech community. Gary Marcus, a professor emeritus of psychology and neural science at New York University, wrote in his article that the battle for AI hegemony is over, at least for now, and the U.S. has not won. He pointed to three impressive articles published by two Chinese companies in the past few weeks, and the release of DeepSeek's nearly state-of-the-art model at end-December 2024, which costs one-fiftieth as much to train as previous models, putting DeepSeek on par with OpenAI, Google and Anthropic in terms of performance and innovation. A few weeks later, TikTok's parent company, ByteDance, dropped a bombshell by announcing a newer and cheaper model to wrestle with OpenAI's o1 model. A company based in Hong Kong also released a viable model, though not as powerful as the DeepSeek’s, requires less training data. The article stated that none of this means that China has won the AI race or is already in the lead, and that U.S. firms will consolidate the latest results and continue to generate new ones on its own. But the fact that China and the U.S. are rapidly moving to a tie is impressive, and the fact that China is able to achieve such good results without tens of thousands of Nvidia (NVDA.US)’s H100 chips is impressive. The rest of the world will catch up, Marcus opined, citing LLMs will be much cheaper, which will lead to the eventual elimination of the need for special-purpose hardware to a certain extent. AAStocks Financial News |
||||
Copied to Clipboard
 |
||||
