I find code (but not all code; there is a difference between environments and languages) as a tools and means of thought to be a lot more efficient than english. I use mostly code to talk to gpt4 and it fills gaps that I really don’t like to fill because the language used (usually typescript these days: I don’t work in a vacuum) is so horribly inefficient in expressing ideas in a fluid-as-thought way.
The ambiguous human language is a pretty horrible vehicle compared to that and I see english to code in LLMs more like the way people use Excel; just keep messing around until it works (with often with inefficient and completely wrong sub parts because the direct and correct solution was never found) without really caring if the end result is really what you meant, because actually you cannot translate what you meant in anything formal like a precise formula or a piece of elegant code.
I see nothing really wrong with that; actually, I have a perverse interest in it; my company troubleshoots and fixes urgent code and infrastructure issues for companies, no matter how bad/old/messy/broken it is. I see my company grow a lot the coming years because of this new wave of software coming up. Because of course, this will go from
> you won’t intersect with the new wave of programmers
To
Mostly everything now is written by this new wave of programmers
Why? Because it’s cheaper in the short run. Why hire me for 300$/hr to write something when the dropout with 2 months of experience waiting tables at Wendy’s can use $20/mo chatgpt4 do it faster and for far less? And I agree with that from a business standpoint IF the software wouldn’t need to live long.
However, I am older and because of my company activities, I know that a LOT of these short lived departmental or startup software hacks end up as vital parts holding up the company over the decades.
I read somewhere (I think here on HN) of someone saying ‘but no one expects me to statically type out a script that removes dollar signs from a csv’ or something like that. If you would give me a $ for every time I encounter a script like that, that started with that same mindset, but has been integrated over the years into vital processes and breaks things in unexpected ways because it’s not made to be there, I would have a free high spec laptop every year. There are 100s of these in every spaghetti hell of a company I walk into.
LLMs and the tooling around them will get better; there is my hope that, in a few decades, the LLMs will deliver code for programmers and end users alike that is robust and even proven and 100% test coverage mostly without extra effort on the human side. Then I can retire my company. For now, I see a bright future with exciting new and unexpected horrors (although, to be honest, it cannot get worse than what outsourcing sweatshops produce; we are doing a project now which has 1000s of lines (so far, it is early days) of code which are unreadable (hard to explain, but they are logically wrong mixed by human obfuscation to hide the fact they didn’t understand the tasks they ‘completed’); it almost seems a joke to write code like that).