Thread

I‘m reading this as „don‘t invest a lot of time without thinking about what problem to solve, how it fits into your product, etc.“ There is however a similar antipattern of spending too much time on building an elaborate framework to realize the data and/or model doesn‘t work. twitter.com/sarahcat21/sta…

I‘m seeing the latter especially, when the assumption is the problem is solved and you can just „download it from huggingface“ when in reality you‘d need to adapt it to your data to work well.

But essentially, ML Systems are complex, and focussing on any one part to the exclusion of others will lead to problems.