As of March, artificial intelligence successfully completed 66% of tasks on a computer, compared to 72% for humans, according to Stanford University’s annual report on AI capabilities.
In 2024, neural networks could only perform 12% of tasks in a digital environment.
In the key AI programming test, SWE-bench Verified, performance increased from 60% to nearly 100% in just one year. Significant progress is also observed in other key technical benchmarks.
Researchers noted that large language models could win a gold medal at the International Mathematical Olympiad, yet they struggle to accurately tell time.
The best model, Gemini Deep Think, correctly reads analog clock times only 50.1% of the time. Scientists refer to this issue as the “jagged edge.”
Meanwhile, the performance gap between AI models from the US and China has effectively closed. Since early 2025, American and Chinese solutions have frequently exchanged leadership positions.
Technology Adoption and Investment
The adoption of large language model-based solutions in organizations has reached 88%. Meanwhile, four out of five university students use chatbots.
The average adoption rate of generative AI among the population reached 53% over three years—faster than personal computers or the internet. However, rates vary by country and strongly correlate with GDP per capita.
“The adoption of artificial intelligence is spreading at an unprecedented rate, and consumers are reaping significant benefits from tools they often access for free,” Stanford concluded.
In 2025, global corporate investments in the industry reached $581.7 billion—more than double the previous year’s figure. The US accounted for $285.9 billion, 23 times the volume of private investments in China.
The United States hosts the largest number of AI data centers, with the majority of chips produced by a single Taiwanese factory, experts noted.
Pressing Issues
According to the study’s authors, all current systems designed to measure, manage, and implement AI lag significantly behind the technology itself. Industry safety standards are outdated, and the number of incidents has sharply increased.
“Virtually all leading developers of advanced models report performance results, yet reports on responsible AI metrics remain incomplete. The number of documented AI-related incidents rose to 362 from 233 in 2024,” the authors noted.
The situation is exacerbated by recent studies showing that improving one aspect of responsible AI, such as safety, can worsen another, like accuracy.
Stanford also pointed out another issue. More than 80% of American high school and college students now use AI for academic assignments. However, only half of the country’s middle and high schools have technology-related rules, and just 6% of teachers find them clear.
On the other hand, the number of new PhDs in AI in the US and Canada increased by 22% from 2022 to 2024. Many specialists have taken positions in academia rather than the commercial sector.
Back in March, a Stanford University study highlighted the risks of excessive reliance on AI for advice.