Nicholas Carlini, a security and ML researcher, shares his opinionated framework for doing high-impact research and occasionally winning best paper awards. The post covers four main areas: developing good research taste (choosing problems worth solving), executing research well (failing fast, killing bad ideas, going to unreasonable lengths on experiments), writing papers effectively (singular focus, writing for a specific reader, crafting abstracts/introductions/conclusions), and the role of luck (timing, committee selection, being on the critical path of scientific discovery). Key insights include: taste is the most important skill, ideas are cheap so share them freely, papers should have exactly one idea, conclusions should answer 'so what?' rather than restate the abstract, and many award-winning papers were rejected multiple times first. The overarching message is to focus on having impact rather than accumulating publications or chasing awards.
Table of contents
Coming up with your best-paper-worthy ideaWhile doing the best-paper-worthy researchWhen writing down the best-paper-worthy researchAfter writing the best-paper-worthy researchConclusionSort: