Nandini Roy Choudhury, writer
Synopsis
- DeepSeek’s advancements in AI, demonstrated by a low-cost model, indicate a significant competitive landscape between the U.S. and China in artificial intelligence development, challenging the notion that China is lagging behind.
- Despite DeepSeek’s achievements, many tech executives believe it poses a minor threat to established players like OpenAI and Anthropic, as the market still favors larger AI models and traditional development paradigms.
- Concerns have been raised about DeepSeek’s claims regarding costs and the potential use of data from U.S. models, with experts suggesting that while competition is intensifying, the dominance of larger models remains intact.
Detailed
Top technology executives told CNBC that the technological advancements that Chinese artificial intelligence lab DeepSeek has exhibited show that the game is on when it comes to competition between the United States and China in the field of artificial intelligence.
The executives of numerous big technology businesses told CNBC in a series of interviews that took place at the Artificial Intelligence Action Summit in France that the advent of DeepSeek proves that China cannot be left out as a serious participant when it comes to the creation of artificial intelligence.
DeepSeek shocked global markets with a technical paper that stated that one of its new artificial intelligence models was created with a total training cost of less than $6 million. This is a significant amount of money that is significantly less than the billions upon billions of dollars that are being spent by Big Tech players and Western AI labs such as OpenAI and Anthropic.
DeepSeek’s advanced and low-cost model demonstrates that there is a “very real competition between U.S.-led, small D democratic AI and CCP [Chinese Communist Party] China-led autocratic, authoritarian AI,” according to Chris Lehane, chief global affairs officer at OpenAI. Lehane made this statement to CNBC.
DeepSeek has been subjected to criticism from a number of individuals who have pointed out that the model appears to be censoring sensitive issues. The artificial intelligence assistant app of DeepSeek, for instance, responds to a question concerning the slaughter that occurred in Tiananmen Square in 1989 by saying, “Sorry, that’s beyond my current scope.” We should talk about something else, shall we?
As Lehane was speaking to Arjun Kharpal of CNBC on the sidelines of the Paris AI summit on Monday, he stated that there are just two countries in the world that are capable of building this at scale. If there were only two countries in the world that were capable of producing electricity on a large scale, what would that be like? To put it another way, it is the way you should think about it.
“What DeepSeek really reinforces and reaffirms for us is that there is this very real competition with very real stakes,” Lehane added. “This is something that we found to be very important.”
Nevertheless, the majority of tech executives were of the opinion that despite the fact that DeepSeek’s achievement demonstrates that China is further along in the global AI race than was previously believed, the threat that it poses to OpenAI is still relatively minor for the time being.
“The fight has begun.”
However, DeepSeek claims that their new R1 model, which is an open-source reasoning model, was able to compete with the performance of OpenAI’s own equivalent o1 model. This was accomplished through the utilization of a procedure that was less expensive and required less energy.
As a result, experts began to question the conventional wisdom that has prevailed in the Western world over the past few years, which is that China is lagging behind the United States in terms of artificial intelligence development. This is due to export restrictions that make it more difficult for companies in China to acquire more advanced graphics processing units, also known as GPUs, manufactured by Nvidia.
For the purpose of training and executing artificial intelligence systems, graphics processing units (GPUs) are essential since they excel at parallel processing, which means they can do several computations simultaneously.
The new model that DeepSeek has introduced is “a big deal in showing that the game is on,” according to Reid Hoffman, who is a partner at the venture capital company Greylock Partners and a co-founder of LinkedIn. Hoffman made this statement to CNBC on Monday.
In addition to stating that DeepSeek’s R1 is “a credible, actionable model,” Hoffman stated that “the competition is very much on the move with China.”
After speaking with CNBC, Abishur Prakash, the founder of the strategic advice business The Geopolitical Business, stated that DeepSeek demonstrates that the Western world’s understanding of China is still restricted.
In a phone interview with CNBC, Prakash stated that the position that the United States of America has established as the technological captain of the world is no longer an acceptable belief.
“That is the new status quo now, that the space between the United States and China has narrowed almost overnight — but it hasn’t narrowed overnight, it’s been years of progress,” Prakash added. “This distinction has been made possible by years of progress.”
In addition, he stated that “if there is one thing that the West should take away from this, it is that their understanding of China is extremely limited, and we do not know what is going to happen next.”
There is no significant danger to artificial intelligence in the United States;
yet, major executives in the field of artificial intelligence are not yet convinced that DeepSeek poses any kind of significant danger to the companies of AI laboratories such as OpenAI and Anthropic.
Experts are generally in agreement that DeepSeek’s artificial intelligence advancements have been excellent; yet, there have been questions raised about the startup’s assertions regarding the cost.
According to a report that was published by SemiAnalysis, a semiconductor research organization, last month, it was estimated that DeepSeek’s hardware expenditures have been “well higher” than $500 million over the company’s lifetime. When CNBC attempted to get in touch with DeepSeek, the company did not immediately respond with a statement.
According to the findings of the analysis, DeepSeek incurs large costs for research and development as well as expenses associated to ownership. Additionally, the report discovered that the generation of “synthetic data” for the model to train on would need a “considerable amount of compute.”
DeepSeek may have been able to attain such a high level of performance, according to the opinions of some technologists, because it trained its models on larger artificial intelligence systems in the United States.
“Distillation” is the name given to this method, which entails letting more strong artificial intelligence models evaluate the quality of the responses that are being provided by a more recent model.
It is an allegation that OpenAI itself has alluded to, telling CNBC in a statement that it is analyzing reports that DeepSeek may have “inappropriately” used output data from its models to construct its AI model, a practice that is referred to as “distillation.” OpenAI is currently investigating these reports.
According to what Hoffman said to CNBC, “the majority of the market fear surrounding [DeepSeek] is in fact misplaced.” Furthermore, it is still necessary to use huge models because it was derived from large models.
According to him, “I believe the short answer that everyone should take is: game on — but large models still really matter,” he went on to say.
Despite the fact that DeepSeek challenged the “paradigm that brute force scaling is the only way to kind of build better and better models,” Victor Riparbelli, CEO of Synthesia, an artificial intelligence video platform, told CNBC that the notion that businesses are going to suddenly shift significant amounts of their AI workloads is misguided.
“I still think that when you look at users of these technologies, all the workflows, I think when we look back in three months’ time, I think 0.01% of those is going to be moved to Deepseek from OpenAI and Anthropic,” Riparbelli said. “I think this is going to be the case.”
According to Meredith Whitaker, president of the Signal Foundation, the development of DeepSeek does not significantly move the needle for the business because the momentum of the market is still largely in favor of larger AI models. A non-profit organization known as the Signal Foundation provides assistance for the encrypted messaging software known as Signal.
“This is not something that is going to disrupt the concentration of power or the geopolitical balance at this stage,” Whitaker said in an interview with CNBC. “I believe that we need to keep our eyes on the ball there and acknowledge that the ‘bigger is better’ paradigm, which has not been reduced through efficiency gains in the past, is the primary factor that is driving this concentration,”
Source : CNBC news
