Academics Pushed Out as AI Research Shifts to Private Labs
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Schmidt Sciences AI2050 fellows gathered in Mountain View last week at a convening hosted by the Eric and Wendy Schmidt–funded program, with several researchers citing its GPU funding as a major benefit amid reduced federal scientific funding.
- University AI researchers can't afford the GPUs needed for frontier model training, and Anthropic and OpenAI don't expose Claude or ChatGPT internals — UC Berkeley's Nika Haghtalab compared the situation to biologists being locked out of CRISPR.
- Anjalie Field at Johns Hopkins published a study finding LLMs give less sophisticated responses to prompts phrased in ways more commonly used by women than by men — work she said frontier labs are unlikely to pursue because it could make them look bad.
- Google DeepMind disbanded its AlphaFold team last month, the group behind the Nobel Prize–winning protein structure prediction model, while academics building specialized scientific AI tools say they're overshadowed by the public conflation of "AI" with energy-guzzling LLMs.
- OpenAI's models have solved multiple real mathematics research problems, prompting some AI2050 fellows to worry humans may lack a future in pure math — including one fellow's stated concern about the mental health of mathematician peers.
Why it matters: The center of AI research has migrated behind closed doors at frontier labs, and the AI2050 program's GPU funding can't fully compensate — independent work that exposes LLM bias or builds non-LLM scientific AI now depends on resources universities no longer control.
Ask SkimNews
