OpenAI Scholars 2020: Final projects

OpenAI wrapped up its third Scholars cohort this week, with participants presenting final projects at a virtual Demo Day after five months of mentored research. The program, now in its third run, pairs newcomers to machine learning with OpenAI staff and gives them time, compute, and guidance to chase a research question from scratch to something presentable.

That structure matters more than it might seem. Five months is a strange middle length — long enough to move past tutorials and toy examples, short enough that scholars have to make real choices about scope. Demo Day is the payoff: a chance to show peers and mentors what came out of that stretch, warts and all, rather than a polished paper months later.

OpenAI has been vague in public about who exactly made up this class or what topics they tackled, at least in what's been shared so far. But the pattern across prior cohorts gives a decent guess at the shape of things: people from outside traditional ML backgrounds, often self-taught or coming from adjacent fields, spending months getting hands-on with the kind of research infrastructure that's normally locked behind a PhD program or a research lab badge.

The Scholars program has always been sold as part of OpenAI's answer to a diversity problem the whole field has struggled with. Access to serious ML research still runs through a narrow set of universities and companies, and mentorship pipelines like this one are one of the few concrete levers organizations have pulled to change that. Whether five months and a demo day meaningfully dents that problem is a bigger question than any one class can answer, but it's the kind of program that only works if it keeps happening, year after year, without fanfare.