New Stanford report highlights the potential, costs, and risks of AI

With AI occupying center stage for the past few months, the AI Index is a valuable resource to see what the future holds.

What does the AI Index say?

To begin with, the report highlights the dramatic shift in the release of AI models. While until a few years ago, it was academia that led the development of large models, now it is the industry that is leading the charge.

Not only are the industry-led models much bigger, churning a million times more data than what academia did a decade ago, but the sheer number of models released also shows that the industry is significantly ahead. In 2022, compared to three models released by academics, the industry unveiled 22 models.

Additionally, these models have become far more capable than they were only two years ago. To gauge their overall performances, the report states that more comprehensive benchmarking suites need to be released.

The report also found that more businesses are rooting for AI and want to implement it in their workflows to improve employee productivity. Those who have already incorporated AI have reported lowered costs and improved revenues too.

Companies are looking for AI-related skills, but the shift has been much slower on the ground than it appears in the media. Between 2021 and 2022, AI-related job posts increased marginally from 1.7 percent to 1.9 percent.

AI-powered tools like Co-pilot have resulted in increased productivity for employees too, who expressed their ability to focus on more satisfying work with the aid of AI tools.

Even though the rise of AI has been shown to accelerate scientific progress, private funding in the area has actually been reduced. Year-on-year, global investments into AI dropped by a whopping 26 percent in 2022, with a lesser number of AI companies being funded.

This also comes at a time when the costs of developing AI models have shot up significantly. OpenAI trained its GPT model in 2019 for an estimated amount of $50,000. Google's large language model PaLM cost the company $8 million last year.

Those interested in reading the full 386-page report can find it here.