【By Observer Net, Ruan Jiaqi】

After the release of OpenAI's new model GPT-5, it faced a "disaster," and CEO Sam Altman could not hide his disappointment.

Speaking to reporters, he admitted, "It shocked me... we completely messed up."

Just an hour ago, under the pressure of the fierce criticism, OpenAI had to issue an update, reverting ChatGPT's default model to the previous version.

Sitting in a Mediterranean restaurant in San Francisco, meeting with journalists, Altman still carried the frustration of dealing with the mess, "I think we have learned our lesson and understood what it means to upgrade products for hundreds of millions of people in one day," he said.

But this setback did not prevent him from starting a new topic, continuing to spread the so-called "Chinese AI risk."

According to U.S. media outlets such as CNBC, Altman said in an interview that the United States has misjudged the issue of China's AI development, possibly underestimating its complexity and severity.

"I am worried about China's (AI development)," Altman warned, saying that the U.S.-China AI competition is deeply intertwined, and its impact goes beyond a simple ranking of who is ahead.

"For example, in terms of reasoning computing power, China may be able to build it up faster. In addition, there are aspects such as research and product development; this matter involves many dimensions," he said, "I think it's not that simple that whoever leads will win the AI competition."

Although the U.S. export controls on semiconductors continue to be upgraded, Altman remains concerned that this policy cannot keep up with the actual technological developments, believing that export controls alone are not a reliable solution.

Media mentioned that according to a new agreement by the Trump administration, U.S. chip companies NVIDIA and AMD have agreed to pay 15% of their chip sales revenue in China to the U.S. government to obtain export licenses issued by the Trump administration.

Although he did not directly mention this agreement in the interview, Altman emphasized that the U.S. government trying to control the AI development process solely through policy measures is "unrealistic."

He explained that the U.S. focuses only on the single aspect of "blocking advanced chip exports," while China is building a complete AI technology ecosystem, which would make even the U.S. regulatory measures, although continuously upgraded, ineffective.

"You can impose export controls on something, but it might not be the key thing... people may build wafer factories or find other ways around it," he said.

When asked if the reduction in the number of GPUs flowing into China made people feel reassured, he still expressed doubt, "My intuition is that it won't work."

"I also hope for a simple solution, but my intuition tells me it's hard," Altman added.

Face with U.S. tech embargoes, although China has long urged the U.S. to lift export controls, CNN noted that China's reaction was not celebratory but clearly indifferent after Trump unexpectedly approved the sale of NVIDIA's H20 chips to China last month.

CNN reports that this unusual attitude reflects China's commitment to building a self-sufficient semiconductor supply chain, as well as confidence in the progress made in its rapidly developing chip industry. However, despite major advances in the semiconductor field, news that "Chinese self-researched chips have broken through" keeps coming out, but China still has a need for more advanced artificial intelligence (AI) processors affected by U.S. export controls.

Previously, Chinese Foreign Ministry spokesperson Lin Jian responded, stating that China has repeatedly expressed a firm position on the U.S. maliciously blocking and suppressing China's semiconductor industry. The U.S. politicizes, overemphasizes security, and uses trade and technology issues, continuously increasing export controls on Chinese chips, pressuring other countries to suppress China's semiconductor industry. This behavior hinders the development of the global semiconductor industry and will eventually backfire, harming both sides.

Photo of OpenAI CEO Sam Altman

This meal, Altman probably found particularly unpleasant. During the meal, he also admitted that China's AI development forced OpenAI to change its model release strategy.

According to CNBC, although OpenAI has long refused to fully open-source its technology, Altman admitted that competition from Chinese models, especially the emergence of open-source systems like DeepSeek, was one of the important reasons for OpenAI's decision to launch an "open-weight model" (Open-weight Model).

"Clearly, if we didn't do this, the AI ecosystem in the world would most likely be built on Chinese open-source models," Altman said, "This was definitely one of the factors that pushed us to make the decision. Although not the only factor, it had a significant impact."

At the beginning of August, OpenAI launched two open-weight models, gpt-oss-120b and gpt-oss-20b, which were free and supported developers to customize, offering a more transparent alternative to closed-source products.

This was the first time since 2019, when OpenAI released GPT-2, that it launched an open-weight model, and it was also the first such model to be released in six years after OpenAI signed an exclusive cloud service agreement with Microsoft.

U.S. media called this move a "major strategic shift" for the company, which has long managed its technology in a closed manner. OpenAI is betting on "increasing the accessibility of technology" to expand the developer ecosystem and strengthen its advantage over Chinese competitors.

Altman previously commented that OpenAI's past closed model approach was "on the wrong side of history." However, the so-called "open weights" and "open source" (open source) are not the same concept. OpenAI has not yet publicly released its training data or full source code.

Moreover, the debut of the first batch of open weight models received mixed reactions. Some developers felt these models lacked highlights, stripping away many core powerful functions of OpenAI's commercial products.

Altman did not deny this. He said that his team intentionally optimized for a core application scenario, aiming to support people in building local-running programming agents (coding agents).

"If the needs of the future world change, it can also be adjusted for other uses," he said.

This article is an exclusive contribution from Observer Net, and it is not allowed to be reprinted without permission.

Original: https://www.toutiao.com/article/7540136101449712167/

Statement: The article represents the personal views of the author. Please express your opinion by clicking the [Up/Down] buttons below.