"The race for 'AI Supremacy' is over, at least for now, and the U.S. didn't win." It just became a tie with China.

[highplainsdem]

The sentence in quotes in the thread title is from AI expert and critic Gary Marcus's latest Substack, just this morning. Link and excerpts from that below, and links to other articles and a video.

I've been so focused on politics this past week that I gave almost no attention to AI news, or Marcus's tweets and Bluesky posts. I also missed usonian's post about this on Thursday - https://www.democraticunderground.com/10971911 - in the Open Source and Free Software group here. Mea culpa. I should have posted about this in LBN days ago. Will try to explain why this is a BFD now.

Gary Marcus's Substack about this, which I hope you'll read in its entirety:

https://garymarcus.substack.com/p/the-race-for-ai-supremacy-is-over

The race for "AI Supremacy" is over, at least for now, and the U.S. didn't win. Over the last few weeks, two companies in China released three impressive papers that annihilated any pretense that the US was decisively ahead. In late December, a company called DeepSeek, apparently initially built for quantitative trading rather than LLMs, produced a nearly state-of-the-art model that required only roughly 1/50th of the training costs of previous models, — instantly putting them in the big leagues with American companies like OpenAI, Google, and Anthropic, both in terms of performance and innovation. A couple weeks later, they followed up with a competitive (though not fully adequate) alternative to OpenAI's o1, called r1. Because it is more forthcoming in its internal process than o1, many researchers are already preferring it to OpenAI's o1 (which had been introduced to much fanfare in September 2024). And then ByteDance (parent company of TikTok) dropped a third bombshell, a new model that is even cheaper. Yesterday, a Hong Kong lab added yet a fourth advance, making a passable though less powerful version of r1 with even less training data.

-snip-

Realizations about Deepseek are already reverberating. Altman has already cut prices twice in the last few days, and we already know from his own testimony that OpenAI has been losing money on ChatGPT Pro; as Grady Booch observed on X, “Now he’ll be losing money faster.”

Nvidia could soon take a serious hit, too, for two reasons. First, the DeepSeek results suggest that large language models might be trained far more efficiently going forward. Nvidia has basically been getting rich selling (exceptionally well-designed) shovels in the midst of gold rush, but may suddenly face a world in which people suddenly require far fewer shovels.

Second, the DeepSeek results suggest that it is possible to get by without NVidia's top-of-the-line chips. Devices shipped in armored cars and traded on the black market might suddenly seem "nice-to-haves" rather than must-haves.

-snip-

From Investor's Business Daily this morning:

https://www.investors.com/market-trend/stock-market-today/dow-jones-futures-meta-tesla-microsoft-buy-points-earnings-due/

AI infrastructure stocks got a boost last week from the big Stargate project announced by President Trump. But, despite Meta's big capital spending plans, many AI hardware stocks fell Friday, including Nvidia. Why? Chinese AI lab DeepSeek released a powerful AI model that it claims cost just $6 million to build, using less-powerful chips. If true — or partially true — that could undermine demand for super-pricey AI chips and other gear.

Articles:

https://mashable.com/article/deepseek-ai-hands-on

https://venturebeat.com/ai/deepseek-r1s-bold-bet-on-reinforcement-learning-how-it-outpaced-openai-at-3-of-the-cost/

https://www.wired.com/story/deepseek-china-model-ai/

https://www.forbes.com/sites/petercohan/2025/01/26/nvidia-stock-may-fall-as-deepseeks-amazing-ai-model-disrupts-openai/


CNBC video on DeepSeek: