You pretty much just have to play around with them enough to be able to intuit what things they can do and what things they can't. I'd rather have another underling, and not just because they grow into peers eventually, but LLMs are useful with a bit of practice.
I can call something "purple" and then give a specific explanation I mean "computer" by it, but I just shouldn't be using purple like that in the first place. The word "moderate" is too entrenched, gives people an immediate instinctive, emotional reaction based on the established meaning of the word. This is not the word to repurpose here.
It's because their test is bad at testing this behavior, this behavior is useful for a "bursty" traffic pattern. The point is to keep the OS scheduler from expensively context switching out of BEAM when it's idle, requiring an expensive context switch back when a new burst of requests comes in. You'd want to send it a heavy burst of 1000 requests at random delays, or something.
The only time you'd want to disable it is if you're under a CPU quota, or if you have some other important OS process that's not getting enough CPU.
Somewhere in between. There actually is a level of demand for languages which are annoying to use; whether this is annoying syntax or annoying semantics isn't much of a distinction, they dovetail. But more than that, there's demand for ML clones (but with curly braces), and neither ML nor generic-curly-brace is the strongest foundation to build readability on, unfortunately.
