Stanford’s recent report highlights the overall rise of Chinese artificial intelligence (AI), noting that models developed by Chinese companies achieve scores similar to those of their American counterparts on the LMSYS benchmark. The report observes that China leads in publishing AI research papers and filing AI-related patents compared to the United States, though it does not evaluate the quality of these outputs. In contrast, the U.S. is noted for generating more prominent AI models, with 40 significant models compared to China’s 15 and Europe’s three. The document also identifies emerging powerful models from regions such as the Middle East, Latin America, and Southeast Asia as AI technology gains global traction.
The research indicates a trend toward “open weight” AI models, which can be freely downloaded and modified. Meta is central to this movement, having released its Llama model in February 2023, with the latest iteration, Llama 4, also recently launched. Companies like DeepSeek and the French firm Mistral have also developed advanced open weight models. In March, OpenAI announced plans to release an open-source model by summer, its first since GPT-2, revealing that the gap between open and closed AI models decreased from eight percent to 1.7 percent in 2024. Nevertheless, 60.7 percent of advanced models remain closed.
The report from Stanford underscores progress in AI efficiency, citing a 40 percent enhancement in hardware efficiency over the past year, which has reduced the costs of querying AI models and enabled the operation of capable models on personal devices. This increased efficiency has sparked speculation that fewer GPUs might be needed for training the largest AI models. However, most AI developers still require more computing power. The study notes that modern AI models utilize tens of trillions of data tokens and tens of billions of petaflops of computation. It also references research suggesting that internet training data might be depleted between 2026 and 2032, accelerating the shift to synthetic, or AI-generated, data.
The report provides a comprehensive view of AI’s growing influence, stating that demand for machine learning skills has risen, with surveys indicating that more workers expect AI to impact their roles. In 2024, private AI investment hit a record $150.8 billion, with governments worldwide investing billions in AI during the same year. AI-related legislation in the U.S. has doubled since 2022.
Despite the increasing secrecy among companies regarding the development of frontier AI models, academic research is reportedly thriving and improving in quality. The report also highlights challenges associated with AI adoption, noting a rise in incidents of AI misuse or malfunction and an increase in research focused on enhancing model safety and reliability.
Regarding the pursuit of artificial general intelligence (AGI), the report observes that some AI models already surpass human performance on specific skill-based benchmarks, including image classification, language comprehension, and mathematical reasoning. This progress is partly due to models being engineered to excel in these areas, underscoring the rapid advancements in AI technology in recent years.
