AI professors are negotiating the new realities of academic research

Last week, the author attended a convening of the Schmidt Sciences AI2050 program in Mountain View, California, where many leading AI researchers gathered. The piece notes that in the past four years, AI research has reoriented around large language models, and its cutting edge has moved from academic institutions to private companies. Universities cannot afford the GPUs needed to train frontier models, and Anthropic and OpenAI do not reveal the inner details of systems like Claude and ChatGPT.

Nika Haghtalab, a UC Berkeley CS professor, likened the situation to biologists in a world where private companies had exclusive control over CRISPR. Although AI2050 provides fellows with GPU funding, money remains a pressing concern, especially with reduced federal scientific funding. Even querying commercial models for research purposes can be prohibitively expensive. Consequently, many fellows focus on questions unlikely to be addressed by tech companies. Anjalie Field from Johns Hopkins said she tries not to work on problems a tech company will solve, noting that profit-driven firms may avoid research that makes them look bad. She gave an example: her study found language models give less sophisticated responses to prompts phrased in ways more commonly used by women than by men. The author observes it is difficult to imagine that kind of research coming out of a private company.