Requests for Research 2.0

OpenAI has published a fresh set of seven unsolved problems, calling it Requests for Research 2.0, a sequel to an earlier list the company put out for the same reason: to point outsiders at the parts of AI research that are still genuinely open.

What makes this interesting isn't the list itself, since OpenAI hasn't spelled out granular problem statements in public detail here, but the framing. These aren't corporate PR problems or benchmark leaderboard stunts. They're questions that surfaced organically while OpenAI's researchers were working on other things, the kind of loose thread you notice mid-project and don't have time to pull on. Instead of shelving them, the company is handing them to anyone curious enough to try.

This is very much in the spirit of the original Requests for Research, which became something of a rite of passage for grad students and independent researchers looking to get noticed without an insider connection. A handful of well-known names in the ML community trace early credibility back to tackling one of those problems seriously and writing it up. OpenAI seems to be betting the same thing happens again: that talent shows up in solved problems, not just résumés.

And there's a broader signal here too. Even a lab with OpenAI's resources runs into dead ends and puzzles it can't immediately staff internally. Publishing them is a quiet admission that research bandwidth is finite even at the frontier, and that useful ideas can come from anywhere, including someone tinkering alone with a laptop and a GitHub account.