University AI researchers are adapting to a landscape where frontier model development is increasingly dominated by private companies with greater resources.
Key facts
- •The AI2050 program, funded by Eric and Wendy Schmidt, supports academics working in the field of artificial intelligence.
- •Universities lack the resources to train frontier models, and private companies do not share the internal details of their tools.
- •Anjalie Field found that language models provide less sophisticated responses to prompts phrased in ways commonly used by women.
- •Google DeepMind’s AlphaFold team, which developed a Nobel Prize-winning protein structure prediction model, was disbanded last month.
- •Some researchers believe AI tools will make human scientists more efficient rather than replacing them entirely.
Academic AI researchers, many of whom are part of the Schmidt Sciences AI2050 program, are facing significant challenges as the cutting edge of AI development shifts from universities to private firms. High costs for computing power and limited access to proprietary models like ChatGPT and Claude have forced academics to rethink their research focus and funding strategies.
Resource Constraints and Proprietary Models
University researchers face a disadvantage compared to private labs, which control the design and training of frontier models. UC Berkeley professor Nika Haghtalab compared the current academic environment to biologists lacking access to proprietary gene-editing tools. While the AI2050 program provides funding for GPUs, researchers struggle with the high costs of querying commercial models and a general reduction in U.S. federal scientific funding.
Shifting Research Priorities
Many academics are pivoting away from model capability advancement to focus on questions unlikely to be addressed by profit-driven companies. Johns Hopkins professor Anjalie Field noted that she avoids problems likely to be solved by tech firms, instead focusing on research such as identifying gender-based biases in language model responses. Meanwhile, researchers working on specialized AI models for fields like climate science face challenges in distinguishing their work from energy-intensive large language models.
The Future of Academic AI
The academic landscape is evolving, with many researchers taking leaves of absence to join industry or holding dual roles. Some experts express concern over the potential for AI to automate tasks in fields like pure mathematics. However, others remain optimistic, suggesting that AI could increase human efficiency and that resource constraints may drive academics to innovate with smaller, more efficient model architectures.
Advertisement
This article was independently rewritten by ManyPress editorial AI from reporting originally published by MIT Technology Review.
