Stanford University’s latest report: The number of basic AI models in the United States is five times that of China

Estimated read time 11 min read

On April 17, Li Feifei’s team at Stanford University’s Human-Centered Artificial Intelligence (AI) Institute (HAI) recently released the “2024 Artificial Intelligence Index Report.”

This report is more than 502 pages long and is the seventh AI Index report released by HAI. It tracks data and focuses on investment and financing activities in the global AI industry, AI training costs and technological progress, and public perceptions of AI. tidy.

This report has a high influence on a global scale and has been widely quoted by many heavyweight media such as The New York Times, Bloomberg and The Guardian. At the same time, it is also an important reference for policymakers in many countries such as the United States, the United Kingdom, and the European Union, and is one of the most credible and authoritative sources of data and insights in the AI ​​industry.

The latest report points out that in 2023, total global AI investment will drop to US$189.2 billion, a 20% decrease compared with 2022. However, investment in the field of generative AI has surged, increasing nearly 8 times from 2022 (about $3 billion) to $25.2 billion.

The cost of training large models is also becoming increasingly expensive. The report estimates that the cost of training global advanced AI large models has reached unprecedented levels. For example, OpenAI’s GPT-4 training cost is $78 million, and Google’s Gemini Ultra training cost is $191 million.

In terms of the comparison of AI strengths between China and the United States, the United States has obvious advantages in terms of the number of large AI model releases and investment and financing activity.

In 2023, the United States released a total of 109 basic models, more than five times that of China (20). At the same time, the United States has 61 well-known large AI models, while China has only 15.

At the same time, the investment in the AI ​​industry in the United States in 2023 will reach US$67.2 billion, which is 8.7 times that of China (US$7.8 billion).

In addition, China occupies a global dominant position in the number of global AI patents and industrial robot installations. China’s number of AI patents in 2022 will account for 61.1% of the world’s total, far exceeding the United States’ 20.9%. In 2022, China’s installed industrial robots will be 290,300 units, 7.4 times that of the United States (39,500 units).

The following are the ten key points of the report compiled by TMTpost:

1. AI beats humans at some tasks, but not at all

AI has surpassed human performance on multiple benchmarks, such as image classification, visual reasoning and English understanding. However, AI still lags behind humans in more complex tasks such as competition-level numbers, visual common sense reasoning and planning.

2. Industry continues to dominate cutting-edge AI research

In 2023, 51 large influential machine learning models emerged in the industry, while academia only contributed 15. Industry and academia also collaborated to contribute 21 large and influential models in 2023, with the number hitting a record high.

3. The cost of training cutting-edge large models becomes more expensive

It is estimated that the cost of training the most advanced AI large models has reached unprecedented levels.

Among them, OpenAI’s GPT-4 training cost is US$78 million, and Google’s Gemini Ultra training cost is US$191 million. Interestingly, Google’s Transformer model in 2017 introduced an architecture that supports almost all major language models today, and the training cost is only $930.

4. The United States surpassed China, the European Union and the United Kingdom to become the main birthplace of the world’s top AI large models

In 2023, a total of 61 well-known large AI models appeared in the United States, far exceeding the 21 in the EU and 15 in China.

5. There is a serious lack of reliable and standardized assessment of large model liability.

New research shows that there is a serious lack of standardization for responsible AI models. Cutting-edge developers like OpenAI, Google, and Anthropic test their large models against different responsible AI benchmarks, a practice that complicates efforts to systematically compare the risks and limitations of top AI large models.

AI Index pointed out that it is difficult to distinguish the authenticity of existing AI deep fake content, especially political deep fakes that have affected elections around the world.

6. Investment in generative AI surges

Although private investment in AI declined overall in 2023, the amount invested in generative AI surged to $25.2 billion, a nearly nine-fold increase from 2022. Companies including OpenAI, Anthropic, Hugging Face, and Inflection AI have all received significant rounds of financing.

In 2023, the number of generative AI startups that announced successful financing was 99, an increase of 76.79% from 56 in 2022.

7. AI improves human work efficiency and work quality

In 2023, some studies show that AI can help humans complete work tasks faster and improve the quality of their work. In addition, AI can also bridge the skills gap between low-skilled and high-skilled workers. However, some studies warn that using AI without proper supervision can lead to performance degradation.

8. AI further accelerates scientific development

In 2022, AI will begin to drive scientific discovery. More important science-related AI applications will emerge in 2023 – AlphaDev to improve the efficiency of algorithm sorting and GNoME to facilitate the material discovery process.

In recent years, AI has achieved significant improvements on the MedQA (Medical Question Answering Dataset) benchmark. In 2023, the accuracy of GPT-4 Medprompt has reached 90.2%, an increase of 22.6% from the highest score in 2022, and a full 2 ​​times improvement in performance compared to 2019.

9. The number of U.S. AI regulations has increased dramatically

AI-related regulations in the United States have increased significantly over the past year and even over the past five years. In 2023, the United States introduced a total of 25 AI-related regulations, a year-on-year increase of 56.3%, while there was only one in 2016.

10. People around the world are becoming more aware of the potential impact of AI

A survey by Ipsos, a global market research organization, showed that last year, the proportion of people who believed that AI would greatly affect their lives in the next three to five years rose from 60% to 66%. In addition, 52% of people expressed concerns about AI products and services, an increase of 13 percentage points from 2022.

According to the Pew Research Center, 52% of people in the United States say they are more worried than excited about AI, up from 37% in 2022.

Competition between China and the United States in AI strength

In the world-class AI competition, China and the United States are the two countries that attract the most attention. Both sides have their own advantages in different fields of AI.

The report shows that in terms of the number of large AI models, the United States released a total of 61 well-known large AI models in 2023, far exceeding the EU’s 21 and China’s 15, and is the main birthplace of the world’s top AI large models.

The United States also leads the way in base models. In 2023, the United States released a total of 109 basic models, more than five times that of China (20).

In terms of AI investment and financing, the United States also has obvious advantages.

In 2023, the investment in the AI ​​industry in the United States will reach US$67.2 billion, which is 8.7 times that of China (US$7.8 billion), the second largest investor. Since 2022, private investment in AI has fallen by 44.2% and 14.1% respectively in China and the EU (including the UK), while the US has experienced significant growth of 22.1% during the same time.

In the ten years from 2013 to 2023, the United States’ total investment in the AI ​​industry reached US$335.2 billion, followed by China at US$103.7 billion.

In private investment in generative AI, the gap between China and the United States is even more obvious. In 2023, the United States’ total investment in the field of generative AI will be US$22.46 billion, while China’s will be only US$650 million.

Looking at the subdivisions of AI private investment, the three most popular areas in 2023 are AI infrastructure/research/governance (USD 18.3 billion), NLP and customer support (USD 8.1 billion), and data management and processing. ($5.5 billion).

Investment in AI infrastructure/research/governance is dominated by the United States. However, in the field of facial recognition, China’s total investment in 2023 (US$130 million) exceeds that of the United States (US$90 million). In the field of semiconductors, China’s total investment ($630 million) is almost the same as that of the United States ($790 million).

In 2023, the number of AI start-ups in the United States that announced successful financing was 897, and that in China was 122. In the past ten years (2013-2023), the number of American AI start-ups that have successfully raised funds is 5,509, which is 3.8 times that of China (1,446).

Although the United States leads the way in terms of basic model research and development and AI investment and financing activity, China occupies a global dominant position in the number of global AI patents and industrial robot installations.

From 2021 to 2022, the number of global AI patent authorizations has increased significantly by 62.7%. Among them, China’s number of AI patents in 2022 will account for 61.1% of the world’s total, far exceeding the United States’ 20.9%.

China’s demand for industrial robots has grown rapidly in the past decade. In 2013, China’s industrial robot installations accounted for 20.8% of the global total. By 2022, this proportion has increased to 52.4%, ranking first in the world.

Data shows that the number of industrial robots installed in China in 2022 will be 290,300 units, which is 7.4 times that of the United States (39,500 units). The industry with the largest installed base of industrial robots in China is the electrical/electronics industry, followed by the automotive industry and the metal/machinery industry.

As of 2022, the United States leads the field of professional service robot manufacturing, with 218 manufacturers, approximately 2.06 times that of China (106).

The performance of closed-source large models is better than that of open source, and Google ranks first in the industry in terms of the number of large models

In 2023, relevant organizations around the world released 149 basic large models, twice the number released in 2022. 65.7% of these newly released large models are open source, up from 44.4% in 2022.

However, the performance of closed-source large models is still better than that of open-source large models. In multiple benchmark tests such as mathematical reasoning, coding ability, agent behavior, and multi-language understanding average, the closed-source large model achieved a median performance advantage of 24.2%. On the AgentBench agent task, the performance difference between the closed-source large model and the open-source large model is as high as 317.7%.

According to statistics from Stack Overflow, a Q&A website for programmers, the most popular AI development tools among professional developers in 2023 are GitHub Copilot (56.04%), Tabnine (11.74%), and AWS CodeWhisperer (4.91%); the most popular AI search tool The order is ChatGPT (83.3%), Bing AI (18.8%), WolframAlpha (11.2%); the most popular cloud platforms are Amazon Web Services (53.08%), Microsoft Azure (27.80%), and Google Cloud (23.59%) .

It is worth noting that as AI models bloom, the computing power consumed behind them is increasing exponentially. For example, Google Gemini Ultra requires 50 billion PetaFLOPs (1 PetaFLOP equals 1 quadrillion floating-point operations per second) computing power during training, ranking first among all well-known models.

Google is also the company that releases the largest number of large models in the industry. In 2023, Google released a total of 18 large models, including Gemini and RT-2, followed by Meta and Microsoft, which released 11 and 9 respectively.

More and more companies are using AI to empower their businesses. The report shows that 55% of organizations use AI in 2023, up from 50% in 2022 and 20% in 2017. AI application scenarios mainly appear in automated contact centers, personalized customized content, and acquisition of new customers.

In addition, AI Index also conducted a global survey on AI attitudes, with a sample of 22,816 adults (16 to 74 years old) from 31 countries. Among them, more than half believe that AI will change their jobs, while more than one-third believe that AI will replace them.

Specifically, 66% of Generation Z (those born between 1995 and 2009) respondents and 46% of Baby Boomer (those born between 1946 and 1964) respondents believe that AI will significantly impact their current work. At the same time, respondents with higher incomes, higher education, and decision-making positions believe that AI will have a huge impact on their employment.

Broken down by country, 69% of Australians and 65% of British people answered yes to the question “Do AI products and services make you nervous?” Japan has the lowest level of concern about AI products, at 23%.

Although the impact of AI on the job market has caused public concern, AI Index cited a research report released by Goldman Sachs in 2023 and pointed out that AI will increase global annual productivity by 1.0% to 1.5% in the next decade.

You May Also Like

More From Author

+ There are no comments

Add yours