Slightly related, in ML I write a lot of code which will be executed exactly once. Data analysis, creating a visualization, one-off ETL tasks.

There are a lot of times where I could spend mental energy writing “correct” code which trades off space for time etc. Sometimes, it’s worth it, sometimes not. But it’s better to spend an extra 30 seconds of CPU time running the code than an extra 10 minutes carefully crafting a function no one will see later, or that someone will see but is harder to understand. Simpler is better sometimes.

What Big O gives you is an ability to assess the tradeoffs. Computers are fast so a lot of times quadratic time doesn’t matter for small N. And you can always optimize later.