> In the scientific world, he says, it may be risky if researchers actively choose to study trendy topics that hit the headlines, but aren’t as important as other pressing issues that demand more time and effort.
I think scientific research has had its own trends for a long time now, though they probably are more local than what the author is writing about.
I can recall a conversation I had with an acquaintance at an APS fluid dynamics conference last year about how trendy the broad field of fluid dynamics can be. 5 years ago it was all about "dynamic mode decomposition" (which hasn't died yet but is on the wane) and now it's machine learning.
You can be confident a-priori that most machine learning studies in fluid dynamics aren't going to do much of value because their "training data" is deficient in some way, whether the number of data points or the variation of an important variable. (There are other problems as well but that one seems most accessible to HN folks.) These more fundamental issues are not well appreciated, probably because they are hard and not sexy.
It's a lot easier and more exciting to try $buzzword for $X. (If you've missed the trend you always could try comparing $buzzwordA and $buzzwordB for $X. One talk at a conference I'm going to next week fits this template!)
I think scientific research has had its own trends for a long time now, though they probably are more local than what the author is writing about.
I can recall a conversation I had with an acquaintance at an APS fluid dynamics conference last year about how trendy the broad field of fluid dynamics can be. 5 years ago it was all about "dynamic mode decomposition" (which hasn't died yet but is on the wane) and now it's machine learning.
You can be confident a-priori that most machine learning studies in fluid dynamics aren't going to do much of value because their "training data" is deficient in some way, whether the number of data points or the variation of an important variable. (There are other problems as well but that one seems most accessible to HN folks.) These more fundamental issues are not well appreciated, probably because they are hard and not sexy.
It's a lot easier and more exciting to try $buzzword for $X. (If you've missed the trend you always could try comparing $buzzwordA and $buzzwordB for $X. One talk at a conference I'm going to next week fits this template!)