An Unofficial Guide to Prepare for a Research Position Application

Sakana AI just put out something a little unusual for a frontier lab: a public, unofficial guide to how it evaluates research candidates. Written by Stefania Druga, Luke Darlow, and Llion Jones, it's less a checklist of skills and more a distillation of patterns the trio noticed after interviewing dozens of applicants for research roles.

The headline argument is simple but pointed. Plenty of candidates can spin up a complex system using whatever popular framework is trending that month. Far fewer can walk through why they made each specific design choice, where the approach breaks down, or how they'd rebuild it with more time and better judgment. That gap, according to the guide, is exactly what separates a merely good application from a great one.

What's notable is that Sakana frames this as more than interview prep. The same qualities they're screening for, questioning assumptions, evaluating tradeoffs critically, explaining reasoning clearly, are presented as the actual substance of research itself, not just a hoop candidates jump through to get hired. In other words, the interview isn't testing a separate skill from the job. It's testing the job.

There's an implicit critique buried in here too, aimed at how AI hiring has evolved. With tools like copilots and fine-tuned templates making it trivially easy to produce a working demo, the bar for what counts as impressive has shifted. Sakana's guide is essentially saying: we've stopped being impressed by output alone. Show us the thinking behind it.