06/28 2024 427
OpenAI announced in the early morning of June 25 that it would block API traffic from countries and regions that do not support its services starting from July 9, and China is also on the list of banned countries.
Upon hearing the news, domestic large AI models immediately responded by introducing corresponding "relocation" or "migration" plans. Some even proposed a token giveaway plan equivalent to OpenAI's usage scale (with no upper limit), which was jokingly referred to in the industry as "now Chinese AI can achieve token freedom".
We know that overseas restrictions on Chinese AI have always existed. However, previous bans on AI mainly targeted high-performance AI computing cards from NVIDIA and AMD, while OpenAI's strong ban this time has let the "other shoe drop" on the AI software algorithm level.
From hardware to software, the increasingly expanding scope of bans and increasingly strict restrictions constantly remind us that blocking China's progress in AI, a key technology field, has become an open card.
Faced with this irreversible trend of AI blockade, how much impact will Chinese enterprises be affected? Is China ready for the full localization of AI?
Abandon the illusion: Who is affected by OpenAI's ban?
Since the launch of ChatGPT, OpenAI's API has been open to nearly 190 countries and regions, excluding China. However, some domestic enterprises and users have been able to continue using OpenAI's services through technical means.
OpenAI is not unable to detect traffic from China, but may have previously "raised the gun barrel by an inch".
And just on June 22, the US Treasury Department issued a draft rule further restricting US individuals and companies from investing in China's semiconductor, quantum computing, and artificial intelligence businesses. With the introduction of the new draft rule and facing an increasingly clear AI blockade situation, OpenAI also took the initiative to protect itself by strengthening regional restrictions and taking additional measures to block API traffic from unsupported regions.
So who and what companies are "knowingly doing the impossible" and using OpenAI's API? There are mainly three categories:
First, some self-developed AI model factories. Some model factories will call OpenAI's API during the research and development stage, using its GPT products for model training, data comparison iterations, etc. In fact, the training of Google's Gemini-Pro large model also used data generated by Baidu Wenxin. Previously, a domestic internet company was exposed for often reaching the maximum access limit of OpenAI's API, but the company also stated that it only used OpenAI's API during the initial exploration stage at the beginning of the year and stopped doing so in April this year.
Second, shell AI companies. Some start-ups can quickly launch AI products or services by encapsulating OpenAI's API through technical means and pushing it to the market as their own product. In fact, every interaction of users will complete through API calls to OpenAI's models.
Third, application developers targeting overseas markets. In the countries and regions supported by OpenAI, they choose OpenAI's API to "stand on the same starting line" as overseas developers.
Currently, the degree of impact on the above groups from OpenAI's ban is not high.
As domestic model factories' models are basically perfected, there is no need to collect data through API calls anymore. The development of overseas applications often requires in-depth understanding of the local market, so the number of domestic developers is also relatively small. In contrast, start-ups with "shell APIs" may suffer the most significant blow, but by "relocating" to domestic large models and quickly finding替代方案 with similar capabilities, they can also avoid risks to a certain extent.
So overall, OpenAI's stricter API restrictions will not cause much turmoil to Chinese AI.
But this does not mean that Chinese AI can rest assured. From the "NVIDIA ban" to the "OpenAI ban," a clear signal has been sent: once the "Pandora's box" is opened, it will not be closed, and the blockade against Chinese AI is unlikely to be withdrawn in a short time.
It's time to abandon the lucky psychology and illusion of "raising the gun barrel by an inch." Facts have proved that the gun barrel can be pulled down and triggered at any time.
Recognizing the reality: What cards can be played against the irreversible AI blockade?
In terms of the intensity of the blockade, the actions of the US government and AI companies are constantly strengthening; in terms of the breadth of the blockade, from the underlying computing power of high-performance AI chips to the underlying algorithms of large models, the "undermining" blockade is gradually extending to every key part of AI infrastructure.
So, after the embargo on computing power and the ban on algorithms, what cards can overseas players play to block China's AI development? Let's sort out the AI software infrastructure: