There's no reason this machine would be good at difficult tasks - they are usually somewhat bespoke, require true creativity and understanding of complex requir

There's no reason this machine would be good at difficult tasks - they are usually somewhat bespoke, require true creativity and understanding of complex requirements, and oftentimes you'll have no example on the internet - normal programming things. However, there's many reasons it would be good at trivial tasks, and arguably most programmers spend most of their time on said trivial tasks, its just the count of them that's overwhelming; switching between multiple different languages, looking up syntax (at least for me!), and deciding on what's a sensical solution for the use case. Nobody cares about the performance characteristics and just wants it to work. So anyway, I don't think a blog post about how you build a new algorithm to come out of any of the LLMs of today, but that doesn't mean I find them underwhelming.