We published an unofficial guide on what we look for when interviewing research candidates at Sakana AI.
This guide is written by Stefania Druga, Luke Darlow, and Llion Jones, and has emerged from interviewing dozens of talented candidates and noticing patterns in what separates good applications from great ones.
Read it: https://pub.sakana.ai/Unofficial_Guide/
Summary
The core principle? Understanding over implementation.
Many candidates can build complex systems using modern tools. Far fewer can explain why each design choice was made, what the limitations are, and how they’d improve it given more time.
Key insights from the guide:
- Ask questions that distill the problem space – Strong candidates don’t just answer questions; they identify core uncertainties and attack them directly.
- Prototype what matters – Build the thing that tests your riskiest assumption, not everything.
- Go deep, not broad – A single unconventional, well-motivated idea beats five standard tweaks.
- Communicate clearly – Lower ambiguity. State conclusions first, identify what you don’t know, stop when the question is answered.
- Balance engineering and creativity – “Good enough” is often good enough. Know when to stop perfecting in favor of demonstrating results.
This isn’t just about interviews—it’s about how we think about research itself. The ability to question assumptions, evaluate critically, and communicate precisely are fundamental to pushing AI research forward.
Read the full guide: https://pub.sakana.ai/Unofficial_Guide/