The article reflects profound changes in AI research at universities as the AI race shifts to the corporate sector.
Over the past four years, AI research has heavily focused on large language models, shifting leadership from academia to companies like OpenAI and Anthropic.
Many universities lack the funding to purchase GPUs to train advanced models, and they also cannot access the technical details of Claude or ChatGPT.
Professor Nika Haghtalab compares the current situation to the field of biology if only corporations were allowed to use CRISPR gene-editing technology.
The AI2050 program by Schmidt Sciences provides funding to purchase GPUs, but the cost of using APIs from OpenAI, Anthropic, and Google remains very high for researchers.
Many scientists have shifted to researching issues that corporations pay less attention to because they are difficult to monetize.
Professor Anjalie Field discovered that language models give less sophisticated answers to phrasing commonly used by women, demonstrating the critical role of independent research.
Many AI researchers do not develop large language models, but instead build specialized AI for climate, science, and physical system simulation.
Experts worry that the public equates AI with energy-guzzling large language models, diminishing interest in other AI directions.
An increasing number of professors are temporarily leaving universities to join AI labs or work concurrently for corporations.
Some mathematicians are worried as AI has solved many real research problems, affecting the morale of the academic community.
Nevertheless, many experts believe AI will help boost research productivity, and resource constraints may drive university labs to create new breakthroughs with smaller, more efficient models.
📌 AI research is strongly shifting to the private sector due to advantages in data, GPUs, and capital. However, academia still plays a vital role in independent research, evaluating model biases, and developing specialized AI to serve science. The author argues that resource constraints themselves could become the driving force for universities to create new, more efficient AI architectures capable of delivering the next breakthrough.
