AI researchers are encountering significant challenges as the field has shifted from academic institutions to private companies. Over the past four years, AI research has reoriented around large language models, with its cutting edge moving away from universities to firms like Anthropic and OpenAI. Universities cannot afford the GPUs needed to train and run frontier models, and even if they could, these companies are not sharing details about their models. This has created a situation where academic researchers are left with limited access to the tools and data that once defined their work.
Many AI academics are adapting by focusing on research questions that private companies are unlikely to address. Anjalie Field, a computer science professor at Johns Hopkins, says she avoids working on problems she believes will be solved by tech companies. She recently conducted a study showing that language models respond less sophisticatedly to prompts more commonly used by women than by men, a finding that seems unlikely to come from a private company. Others are exploring specialized AI models that can analyze data, make predictions, or simulate physical systems, though they face challenges in gaining recognition for their work.
The AI2050 program offers some funding to help researchers buy GPUs, which some say is a major benefit of participating. However, federal scientific funding in the U.S. has decreased, making it difficult for researchers to afford repeated queries to private models. These constraints are reshaping academia, with several prominent academics taking leave from universities to join frontier labs, and many AI2050 fellows holding industry positions alongside their academic roles.
Source: mittr
