The Institute of International Strategy (IISS) of the well -known research institutions released a report recently, pointing out that although the United States restricts high -performance AI chips exporting to China, these control measures may promote Chinese artificial intelligence researchers towards areas with lower calculation requirements and guide them to guide themDevelop new competitive advantages.
According to the report, in recent years, the ability of large language models has been significantly improved. Openai's creation of GPT-3 in 2020 is an important milestone.These improvements are attributed to the creation of larger and more universal model architectures as well as increased dataset size and the amount that technology companies spent to increase the computing power of training models.Empirical studies have shown that there is a close relationship between the size of the data set, the calculation overhead and the parameter count of the given model, and in practice, the calculation overhead is the strongest constraint of model improvement.
In addition, more and more participants are developing large -scale language models and diffusion in multiple dimensions.The research of large -scale language models mainly occurs in the United States, but researchers from other countries, especially China, and specific research institutions in other places, such as DeepMind in the UK, have invested a lot of resources to establish their own models.In addition, the types of institutions that develop language models have expanded to large technology companies such as Google and Microsoft, as well as decentralized researcher collectives.
The report believes that the diffusion of large model technology has two recent impacts on security: these models may be more quality and more content for false information production, and competition for the development of large models may exacerbate the tension of geopolitical Z.
The report also analyzed in October last year, the US government announced that it implemented new export controls on advanced semiconductor chips flowing to China, partly because these chips are crucial to the development of artificial intelligence.Although the established intention of export control is to limit the development of artificial intelligence systems used to monitor or military applications, the language model also highly depends on these advanced semiconductors.The real motivation for implementing export control may be to maintain the advantages of the United States in terms of language models. Whether it is part of the wider artificial intelligence technology competition strategy, or because the government specifically wants to curb China's development in language models.
The report also doubts whether these measures are valid.Although the large language model has been improved due to the improvement of computing power, they cannot continue to do this at the current speed.Researchers are actively seeking methods with higher efficiency development to train similar models.One country attempts to limit the computing power of another country as a means of competitive artificial intelligence development, which may inspire target countries to develop competitive advantages in these more efficient artificial intelligence methods.
In addition, high -quality texts such as books and academic journal articles may soon become a restrictions on the development of language models more urgent than computing power availability.
There may be more and more national development of the development of large models as a national pride, which may exacerbate the competition for its development.The report is worried that as the big model is getting closer to the center of national technology competition, the government may more actively cut off the visit of the language models developed by the residents of the country, thereby further splitting the Internet.