AI Academics Struggle as Frontier Research Moves to Private Labs

👁 112 reads

Academics in the Schmidt Sciences AI2050 program met last week in Mountain View, California, to confront a research environment where the cutting edge of AI has shifted from universities to private companies. The convening, funded by Eric and Wendy Schmidt, drew fellows who collectively represent a wide cross-section of academic AI work - yet the central tension running through the event was one of access and resources.

Nika Haghtalab, a computer science professor at UC Berkeley, compared the situation to biologists living in a world where private companies held exclusive control over CRISPR: researchers can study how ChatGPT and Claude behave externally, but cannot examine their internal design or training, nor influence either. Universities cannot afford the GPUs required to train frontier models, and even repeatedly querying commercial APIs for rigorous study can be prohibitively expensive - a problem compounded by reductions in US federal scientific funding.

Many fellows have redirected their work toward questions private companies are unlikely to pursue. Anjalie Field, a computer science professor at Johns Hopkins, recently found that language models give less sophisticated responses to prompts phrased in ways more commonly associated with women than with men - research that, as she put it, targets “problems that I think are gonna be solved by a tech company.” Meanwhile, researchers outside the LLM space face separate pressures: Google DeepMind’s AlphaFold team, behind the Nobel Prize-winning protein-structure model, was disbanded last month.

// read moreOpenAI Claims Navier-Stokes Breakthrough, but Accusations of Academic Scooping FollowMagazineOpenAI Private Intelligence: ZDR, PSP & GDPR for EnterprisesNewsJeff Dean and Senior Google AI Figures Move On in Shake-UpMagazine

Kinetyra