AI professors are negotiating the new realities of academic research

Eric and Wendy Schmidt are funding a group of academics to keep them from going bust, as universities can no longer afford the hardware needed to train large language models.

Last week I travelled to Mountain View, California, to speak with the Schmidt Sciences AI2050 fellows. The gathering was a mix of media training and roundtable interviews. The list of participants includes many of the most prominent figures in the field, though not everyone could attend. Every time I turned a corner, I saw a scientist I had interviewed before or whose work I respect. I received a science communication award from Schmidt Sciences in 2024.

University researchers face a difficult situation. Over the last four years, the focus of AI research has shifted entirely to large language models, moving the frontier from academic institutions to private companies. Universities simply cannot afford the GPUs required to train and run these systems. Even if they could, companies like Anthropic and OpenAI do not allow outsiders to see the internal details of their models.

Nika Haghtalab, a computer science professor at UC Berkeley, compared the current state of AI academia to biologists operating in a world where private companies hold exclusive rights to the CRISPR gene-editing tool. Experts outside these labs can observe how ChatGPT and Claude behave, but they cannot study the design and training processes in detail, nor can they steer that development.

The AI2050 program provides some funding for GPUs, which several researchers called a major benefit. Money remains a pressing concern, especially with reduced federal scientific funding in the United States. Even researchers who do not run local models face high costs when repeatedly querying OpenAI, Anthropic, and Google systems to study them rigorously.

Many fellows now focus on questions unlikely to be addressed by tech companies. “I try not to work on problems that I think are gonna be solved by a tech company,” says Anjalie Field, a computer science professor at Johns Hopkins. Companies need to make money, so research questions with little profit promise may not be worth investing in, especially if the answers might make the companies look bad. Recently, Field conducted a study finding that language models give less sophisticated responses to prompts phrased in ways more commonly used by women than by men. It is difficult to imagine such research coming from Anthropic or OpenAI.

There is also a large group of AI academics who do not work with large language models at all. Many build specialised AI models that analyse data, make useful predictions, or simulate entire physical systems. These researchers are not necessarily competing with frontier labs, yet they face their own challenges. At the convening, several voiced concerns about how the widespread ignorance of non-large language model AI is affecting their work. Researchers building tools to help address climate change, for example, sometimes struggle to advocate for their work when so many people believe that “AI” means “energy-guzzling large language models.”

All these challenges are changing the landscape of academia. Several prominent academics have recently taken leave from their universities to join frontier labs, and many AI2050 fellows hold industry positions alongside their academic jobs. In the past six months, yet another threat has emerged. OpenAI’s models have solved a number of real research problems in mathematics, and some experts are worried that humans might not have a future in pure math. One fellow I spoke with said she was concerned about the mental health of her mathematician peers.

But it is not all doom and gloom. Empirical science may prove much more difficult to automate than mathematics because collecting data is an intrinsically slow process. Some researchers see AI mathematicians and scientists as a boon rather than a threat. Tim Dettmers, a computer scientist at Carnegie Mellon who works to make AI models faster and cheaper to run, says AI scientists will not replace humans. On the contrary, they could make human scientists far more efficient, giving him and his peers the chance to pursue all the wild and inspired ideas they might otherwise never have time for.

Scientists are a resilient sort. The very resource constraints that prevent them from training frontier models also push them to discover new ways to make models smaller and more efficient, or to explore completely new architectures. If the next big AI breakthrough comes not from a major company but from a scrappy academic lab, I will not be shocked.

What it means

The shift of power to private companies forces academics to either find niche questions tech firms ignore or build smaller, more efficient tools. The barrier to entry for hardware means the next major innovation could come from a small university lab rather than a tech giant.