Top 10 Takeaways from Stanford’s AI Index Report 2024

5/5 - (1 vote)

A decade ago, AI systems struggled with image classification, language comprehension, and math problems. Today, advanced systems like GPT-4, Gemini, and Claude 3 exceed human performance on benchmarks and can generate fluent text, process audio, and explain memes.

In 2023, AI’s rapid progress saw the release of numerous new large language models, with two-thirds being open-source. Notably, Gemini Ultra reached human-level performance on the MMLU benchmark, and GPT-4 achieved a 0.96 mean win rate on the HELM benchmark.

Despite a decline in global private AI investment, funding for generative AI surged. AI’s integration into daily life is growing, with increased mentions in Fortune 500 earnings calls and evidence of AI boosting worker productivity.

What are the skills most in demand among the AI job postings in 2023? See here:

What do computer scientists earn? See here:

With these interesting nuggets out of the way, let’s dive into the top 10 takeaways of the new Stanford report on Artificial Intelligence.

Takeaway 1: AI Outperforms Humans in Specific Areas

AI excels in tasks like image classification, visual reasoning, and English comprehension, but struggles with more complex tasks such as advanced mathematics, visual common-sense reasoning, and planning.

GPT-4 still outperforms most alternative models in many benchmarks:

The Massive Multitask Language Understanding (MMLU) test is one of the best tests that asseses model performance in 57 subject areas including humanities, STEM, social science. We have reached human baseline performance!

For the average person, this means that while AI can handle certain routine and straightforward tasks, it still requires human intelligence for more intricate problem-solving.

Takeaway 2: Industry Leads AI Research

In 2023, the industry developed 51 notable machine learning models, compared to academia’s 15. Additionally, 21 notable models resulted from industry-academia collaborations, marking a new high.

This dominance means that businesses are at the forefront of driving AI innovation, potentially leading to faster commercialization of AI technologies that can improve everyday life.

Takeaway 3: Rising Costs for Frontier AI Models

The training costs for state-of-the-art AI models are skyrocketing. For instance, OpenAI’s GPT-4 training costs were estimated at $78 million, while Google’s Gemini Ultra cost $191 million.

These escalating costs could lead to higher prices for AI-driven products and services, impacting consumers’ access to advanced technologies.

Takeaway 4: U.S. Dominates AI Model Development

In 2023, U.S.-based institutions created 61 notable AI models, significantly surpassing the European Union’s 21 and China’s 15.

However, China has long granted most AI patents:

For the common person, this indicates that the U.S. remains a global leader in AI innovation, potentially providing Americans with earlier access to cutting-edge AI applications and services.

Takeaway 5: Lack of Standardized Evaluations for LLM Responsibility

There is a significant absence of standardized benchmarks for responsible AI. Leading developers like OpenAI, Google, and Anthropic use different benchmarks, complicating systematic risk and limitation assessments.

This inconsistency means that consumers might face varying levels of safety and ethical standards in AI products. Here’s an initial evaluation of the “trustworthiness” of different models:

However, there are multiple standardized tests for LLM performance (e.g., MMMU) which might be more important for now — given the difficulty of “testing” the ethical standards of an AI model anyways.

Takeaway 6: Surge in Generative AI Investment

Despite a general decline in AI private investment, funding for generative AI surged to $25.2 billion in 2023, nearly eight times the amount in 2022. Companies like OpenAI, Anthropic, Hugging Face, and Inflection reported significant fundraising.

This boom means more innovative AI tools and applications are likely to enter the market, enhancing creativity and productivity in everyday life.

However, it is very interesting to see that we’re not at peak investment. The year 2021 has seen far higher investments in AI:

This might indicate that the amazing AI skills we see in 2024 are just the beginning. We’re not at the peak of a hype phase yet.

Takeaway 7: AI Enhances Worker Productivity

Studies from 2023 show that AI improves worker productivity and output quality. AI also helps bridge the skill gap between low- and high-skilled workers. However, improper use without oversight can decrease performance.

For workers, this means that integrating AI into their workflows can make their jobs easier and more efficient, but it’s essential to use these tools correctly.

Takeaway 8: Accelerated Scientific Progress with AI

AI significantly advanced scientific discovery in 2023 with applications like AlphaDev for efficient algorithmic sorting and GNoME for facilitating materials discovery.

For instance, AlphaDev has already made it in the C++ sorting library — proof that it’s a superior AI-generated sorting algorithm. The impact not only on science but on all computing applications is significant!

Here’s the groundbreaking work on materials research:

Check out my article on the topic:

πŸ‘‰ Google Deep Learning – 800 Years of Human Experimentation in One Discovery

This progress means that breakthroughs in science and technology could occur more frequently, leading to faster advancements in health, energy, and other crucial areas that affect everyday life.

Takeaway 9: Increase in U.S. AI Regulations

AI-related regulations in the U.S. have grown substantially, from one in 2016 to 25 in 2023, with a 56.3% increase last year alone.

For the public, this indicates that the government is taking steps to ensure that AI technologies are developed and deployed responsibly, potentially reducing risks associated with these technologies.

Takeaway 10: Growing Public Awareness and Anxiety About AI

A survey by Ipsos revealed that the percentage of people who believe AI will dramatically impact their lives in the next 3-5 years rose from 60% to 66%. Additionally, 52% express nervousness about AI products and services, up from 37% in 2022. In the U.S., Pew data shows 52% of Americans feel more concerned than excited about AI, up from 37% in 2022.

This growing awareness and concern reflect a need for better public education on AI and more transparency from AI developers to build trust with the public.