AI professors face funding gaps and shifting priorities as frontier labs dominate research

At a recent convening of the Schmidt Sciences AI2050 program in Mountain View, California, AI researchers described a field transformed by the rise of large language models (LLMs) and the concurrent shift of cutting-edge work from universities to private companies. The event, which drew many of the field’s leading academics, underscored the challenges facing university-based AI research, including soaring GPU costs and restricted access to proprietary models like those from Anthropic and OpenAI.

Researchers compared their current position to biologists in a world where private companies controlled exclusive access to gene-editing tools like CRISPR. While academics can study the behavior of models such as ChatGPT and Claude, they cannot access or influence the design and training processes behind them. Some AI2050 fellows noted that the program’s funding, which can be used to purchase GPUs, provides critical support, but overall federal scientific funding in the U.S. has declined, compounding financial pressures.

Many fellows are shifting their research agendas toward questions that for-profit labs are unlikely to pursue, such as the social and ethical implications of AI. Anjalie Field, a computer science professor at Johns Hopkins, described research showing that language models respond less effectively to prompts phrased in ways more commonly used by women, a line of inquiry she said was unlikely to emerge from private labs focused on profitability.

Other researchers are focusing on specialized AI models that analyze data, make predictions, or simulate physical systems, areas less directly in competition with frontier labs. However, they face challenges in advocacy and recognition, as public and institutional attention often fixates on energy-intensive LLMs rather than these alternative applications.

The dominance of private labs has also raised concerns about the future of entire disciplines. Some mathematicians worry that OpenAI’s models solving research problems in mathematics could marginalize human mathematicians, with one fellow describing worries about the mental health of peers in the field. Yet not all perspectives were pessimistic: Tim Dettmers, a computer scientist at Carnegie Mellon, argued that AI tools could enhance human scientific productivity, enabling researchers to pursue ideas they might otherwise overlook.

The constraints on academic AI research—financial, technical, and institutional—are pushing researchers toward innovation in efficiency and novel architectures. While the next major breakthrough may still come from a major company, the resource limitations in academia could spur unconventional advances from smaller labs.