OpenAI Scholars 2019: Final projects

Eight scholars, one demo day, and the quiet end of OpenAI's second Scholars cohort. The program exists to pull people into machine learning who wouldn't otherwise have a clear path in — folks without the traditional CS pedigree or the insider connections that usually grease the wheels in this field. Each scholar spends months being mentored by OpenAI staff, working toward a capstone project, and this cycle's crop presented their work in person at OpenAI's offices.

What's notable isn't any single project here, since the source is thin on specifics, but the structure itself. AI research has a reputation problem: it rewards people who already had access to elite labs, elite universities, elite Slack channels. A cohort-based scholars program with stipends and direct mentorship is a deliberate attempt to route around that, even if it only moves the needle for eight people at a time.

And that's the tension worth sitting with. Eight scholars a cycle is a rounding error against the thousands of engineers OpenAI itself employs or influences, let alone the broader industry hoovering up ML talent. Demo days make for nice photos and blog posts, but the real test is what happens to these eight people a year from now — whether they land research roles, whether they keep publishing, whether the program actually changes who gets to build these systems or just changes who gets a nice line on a resume.

Still, incremental is not the same as meaningless. Pipelines like this compound. The first Scholars class fed into companies and PhD programs; this second class will presumably do the same. If OpenAI keeps running it, keeps funding it, keeps making it a real credential and not just a summer internship with better branding, it becomes one of a small number of legitimate side doors into a field that mostly only has a front door guarded by Stanford and a handful of labs.