Chinese AI startup DeepSeek has gained traction over the past few weeks. It surpassed OpenAI’s o1 reasoning model’s capabilities across math, science, and coding at a fraction of the cost incurred by the ChatGPT maker, translating to 3% of the flagship model’s development cost. For context, the AI startup reportedly trained its R1 V3-powered AI model with $6 million using the reinforcement learning technique.
The ground-breaking milestone has prompted people to hop onto DeepSeek’s hype train, dethroning ChatGPT as the most downloaded free AI app in the United States on Apple’s App Store. However, former OpenAI board member Helen Toner seemingly thinks DeepSeek still has a lot more potential to grow and even dominate the space, but only if the stringent AI chips exportation rule imposed by Biden’s administration in 2023 blocking shipments of advanced chips to China is revoked.
According to Toner, scrapping the rule could lead to a “huge victory” for China and DeepSeek. While speaking to Fortune, the former OpenAI board member indicated:
“I think it’s possible that Nvidia will use this to persuade [President] Trump that the export controls are just holding back U.S. industry and that he should just revoke them.”
Toner’s comments on DeepSeek’s success come amid claims that the AI startup reportedly used Microsoft and OpenAI’s data without authorization, potentially constituting copyright infringement.
Interestingly, Microsoft and OpenAI have been involved in several copyright infringement lawsuits. However, OpenAI CEO Sam Altman argued that copyright law doesn’t categorically prohibit using copyrighted content to train AI models. He admitted it’s impossible to develop ChatGPT-like without privileged data.
The former OpenAI board member is uncertain about President Trump’s administration’s direction on the exportation rules imposed by Biden’s regime. “The big question is whether or not the Trump administration will pick up where the last administration left off.”
During Biden’s last week in office, he announced new critical measures that would further make the imposed exportation rules on China more stringent, including a block on China from bypassing NVIDIA by using Huawei via shell companies and getting TSMC to manufacture the chips. It’s anyone’s guess the approach President Trump will embrace amid the rapid advances in AI. “But they could decide to step in and weaken it,” she added.
While DeepSeek’s R1 development process remains a mystery, it’ll be interesting to see how it competes against leading AI frontier models. This is especially after OpenAI placed a $500 billion bet on Stargate to facilitate the construction of data centers across the United States to bolster its AI advances.
“So far, DeepSeek is acting as a fast follower, not leading the pack,” added Toner. “China is doing everything they can to keep up with the U.S. in AI, and they’re doing well at fast-following. But to imply they’re out ahead of us is clearly wrong.”