Stanford's AI Index Report 2025: 10 Charts to Understand the Current State of Artificial Intelligence
TechFlow Selected TechFlow Selected
Stanford's AI Index Report 2025: 10 Charts to Understand the Current State of Artificial Intelligence
AI optimization technologies continue to advance, but at the same time, the application and misuse of this technology are becoming widespread.
Navigating Web3 tides with focused insightsBy Yuanchuzhou Zhi Xin
The newly released "2025 Artificial Intelligence Index Report" indicates that the field of artificial intelligence is maturing rapidly: while AI optimization technologies continue to advance, applications and abuses of the technology are also proliferating.
Published on April 7, 2025, the report was initiated by Stanford University's Institute for Human-Centered Artificial Intelligence (HAI) and led by the AI Index Steering Committee, composed of interdisciplinary experts from academia and industry.
The report continuously tracks key developments across multiple dimensions in AI, including major annual technological breakthroughs, new benchmark records, investment trends in generative AI, application trends in education, progress in regulatory legislation, and more.
Below are 10 core takeaways from the report:
01. Small Models Achieve Performance Breakthroughs
In 2022, the smallest model achieving 60% accuracy on the MMLU (Multi-task Language Understanding) benchmark was Google's PaLM with 540 billion parameters.
By 2024, Microsoft’s Phi-3-mini model achieved the same performance with just 3.8 billion parameters—reducing model size by a factor of 142 within two years.
02. Dramatic Drop in Model Usage Costs
For models reaching GPT-3.5-level performance (64.8% accuracy) on the MMLU test, the cost per million-token query plummeted from $20 in November 2022 to just $0.07 (Gemini-1.5-Flash-8B) in October 2024—a drop of over 280 times within 18 months. Year-on-year declines in LLM inference pricing across different tasks range from 9x to 900x.
03. Chinese Models Rapidly Catching Up
The U.S. still leads in the number of cutting-edge AI models (40 significant models in 2024 vs. China's 15 and Europe's 3), but Chinese models are quickly closing the quality gap.
Performance differences between leading U.S. and Chinese models on core benchmarks such as MMLU and HumanEval have narrowed from double-digit percentage points in 2023 to mere fractions in 2024. China continues to lead globally in volume of AI research papers and patents.
04. Surge in AI Misuse Incidents
According to the AI Incidents Database, there were 233 harmful AI-related incidents in 2024—the highest on record—an increase of 56.4% compared to 2023. Notable cases include deepfake pornography and chatbots allegedly linked to teenage suicides. While not comprehensive, the sharp rise is alarming.
05. Practical Breakthroughs in AI Agents
AI agents show initial promise. RE-Bench, introduced in 2024, established a rigorous benchmark for evaluating agent performance on complex tasks: in short-term tasks (2 hours), top AI systems outperformed human experts by 4x; however, in long-term tasks (32 hours), humans regained an advantage, outperforming AI by 2x.
Notably, AI has already reached expert-level performance in specific domains (e.g., writing certain code), with higher efficiency.
06. Soaring AI Investment
In 2024, private-sector AI investment in the U.S. reached $109 billion—nearly 12 times that of China ($9.3 billion) and 24 times that of the UK ($4.5 billion).
The gap is even wider in generative AI: U.S. investment exceeded the combined total of the EU and UK by $25.5 billion, up from a $21.1 billion gap in 2023.
07. Enterprises Accelerate AI Adoption
Businesses are increasingly adopting AI. In 2024, the proportion of survey respondents reporting AI use rose sharply from 55% in 2023 to 78%. More significantly, the share of enterprises applying generative AI to at least one business function doubled—from 33% to 71%.
08. Explosion in FDA Approval of AI Medical Devices
The number of AI-powered medical devices approved by the FDA is growing exponentially. Since the first approval in 1995, only six had been approved cumulatively by 2015. By 2023, that number surged to 223—demonstrating accelerating adoption of AI in healthcare.
09. U.S. AI Regulation: Driven by State Governments
With slow progress at the federal level, U.S. states have become the main drivers of AI regulation. In 2016, only one state-level AI-related bill was passed nationwide; by 2023, this number had risen to 49.
In the past year alone, it more than doubled to 131. Although the number of AI bills proposed at the federal level has increased, the actual passage rate remains low.
10. Greater Optimism Toward AI in Asia
Global acceptance of AI varies significantly by region: in China (83%), Indonesia (80%), and Thailand (77%), most people believe AI products do more good than harm; in contrast, optimistic views are held by less than half the population in Canada (40%), the U.S. (39%), and the Netherlands (36%).
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
