"I reckon this is even the case when using it as an interactive encyclopedia".
Yes, that is my experience. I have done some C# projects recently, a language I am not familiar with. I used the interactive encylopedia method, "wrote" a decent amount of code myself, but several thousand lines of production code later, I don't I know C# any better than when I started.
OTOH, it seems that LLMs are very good at compiling pseudocode into C#. And I have always been good at reading code, even in unfamiliar languages, so it all works pretty well.
I think I have always worked in pseudocode inside my head. So with LLMs, I don't need to know any programming languages!
It's even more general than that: LLMs seem to be exceedingly good at translating bodies of text from one domain to another. The frontier models also have excellent natural language translation capabilities, far surpassing e.g. Google translate.
In that sense, going from pseudocode to a programming language is no different from that, or even translating a piece of code from one programming language to another.
Yes, that is my experience. I have done some C# projects recently, a language I am not familiar with. I used the interactive encylopedia method, "wrote" a decent amount of code myself, but several thousand lines of production code later, I don't I know C# any better than when I started.
OTOH, it seems that LLMs are very good at compiling pseudocode into C#. And I have always been good at reading code, even in unfamiliar languages, so it all works pretty well.
I think I have always worked in pseudocode inside my head. So with LLMs, I don't need to know any programming languages!