The latest AI Index Report reveals that the field of artificial intelligence is accelerating at a blistering pace, with computing requirements and costs for cutting-edge AI systems reaching astronomical new heights. The 2024 report, an annual analysis of AI trends and developments, shows the number of AI research publications and patents continuing their steady rise year after year. However, it's the frontier of AI research that is truly taking off at a mind-boggling rate.

In 2023 alone, tech giants like Google, Microsoft, OpenAI and others unveiled a staggering 149 new "foundation models" - enormously complex AI systems trained on vast datasets to master a wide range of tasks. That's more than double the amount released just the year prior. Perhaps more remarkably, nearly two-thirds of these powerful new AI models were open source, allowing researchers and developers around the world to leverage and build upon them freely. This open approach represents a stark departure from the walled gardens of Big Tech just years ago.

"We're seeing an AI renaissance on a scale that defies historical comparisons," said Jacques Pitrat, director of the AI Index project. "The advances we're witnessing now dwarf the 'AI Winters' of decades past. Brilliant minds in labs and garages globally are fueling a positive feedback loop of AI breakthroughs." Of course, such trailblazing achievements don't come cheap. The report estimates that Google's latest Gemini Ultra foundation model required an astonishing $191 million worth of computing power to develop and train. Even OpenAI's vaunted GPT-4 language model consumed $78 million in cloud computing resources.

"We're firmly in an era of 'extreme AI' where today's cutting-edge models are vast in their scale, hungry for exascale computing power, and produced at costs rivaling Hollywood blockbusters," remarked Pitrat. "It remains to be seen if academia and smaller players can keep pace with the resources of tech titans."

While the United States is still the world's unrivaled leader in pioneering flagship AI models, the report signals China's growing influence. China now accounts for over 61% of new global AI patents each year, far outpacing the U.S. at just 20.9%.  

Meanwhile, the foundations of the AI revolution are being built on open source. The report tallied a remarkable 1.8 million public AI software projects on GitHub in 2023 - up a whopping 59% from the prior year.

As AI systems grow more capable with each passing year, one question looms: When might AI researchers run out of real-world data to continue advancing their models? Some concerning evidence suggests major models trained predominantly on synthetic data could begin "hallucinating" outputs distorted from reality. With the frontiers of AI rapidly expanding, the 2024 Index makes one thing clear: The artificial intelligence era is just getting started, bringing wonders and risks yet to be fully understood.

Content Source - https://aiindex.stanford.edu/report/