> That train of thought leads to writing assembly language in ed
Hard disagree, LLVM will always outperform me in writing assembly, it won't just give up and fail randomly when it meets a particularly non-trivial problem, causing me to write assembly by hand to fix it. If LLMs would be 100% reliable on the tasks I had to do, I don't think anyone here would seriously debate about the issue of mental attrition (i.e. you don't see people complaining about calculators). The problem is that in too many cases, the LLM will only get so far and you will still have to switch to doing actual programming to get the task finished and the worse you get at that last part the more your skillset converges to exactly the type of things an LLM (and therefore everyone else with a keyboard) can reliably do.
>The problem is that in too many cases, the LLM will only get so far and you will still have to switch to doing actual programming to get the task finished
The LLM makes mistakes sure and isn't a slam dunk tool like a compiler, but it could still save lots of time and be useful.
Some things are fine to let rot. Nobody should spend too much time learning Vue, React or Laravel, or even nhibernate, entity framework or structuremap, for example.
Such frameworks come and go, the knowledge has little value. Save brain cells for more important, long-lasting things instead. LLM's can certainly help with that.
> Some things are fine to let rot. Nobody should spend too much time learning Vue, React or Laravel, or even nhibernate, entity framework or structuremap, for example.
The way you phrase this you make it sound like an LLM can already solve every possible task you would ever get in Vue, React or Laravel but my entire point is that this is simply not true. As a consequence of this, whenever the LLM fails at a task (which gets more likely the more complex the task is) you will still need to know how Vue, React or Laravel work to actually finish your task but this is the exact knowledge you lose if you spend 80% of your day prompting instead of writing code. The more you rely on the LLM to write your code the more the code you are able to produce converges with the one that the LLM can put out.
Poor phrasing perhaps, my point is that you the human should spend time and effort on the challenging and interesting things, and learning things that have lasting value. Not the boring, tedious and quickly irrelevant things.
Remember knockout.js? script.aculo.us? I do, but wish I didn't so those brain cells could know more SQL and vanilla javascript instead. :)
I also think LLM's are way more useful than you give them credit for. I save hours per week already and I'm just getting started in how to get the most value from LLM's. It's clear to me that my ability to phrase questions and include context matters more than which model I use.
To be clear I'm not talking about the cool-aid promises of one-shotting complex apps here, I mean questions like
"Give me a log4jconfig that splits logfiles per errorlevel and day"
"Look at #locationclass, #customerclass, #shop.css and #customerview and make a crud view for customers similar to locations"
"We are converting a vue app to react. Look at #oldvue1 and #newreact1 and make a react version of #oldvue2 following the same patterns"
"What could cause the form from #somewebview to pass a null whateverId to #somerepository?"
Questions like that are solved by LLM's at least close enough to 100% that it feels like asking a human to do it.
Hard disagree, LLVM will always outperform me in writing assembly, it won't just give up and fail randomly when it meets a particularly non-trivial problem, causing me to write assembly by hand to fix it. If LLMs would be 100% reliable on the tasks I had to do, I don't think anyone here would seriously debate about the issue of mental attrition (i.e. you don't see people complaining about calculators). The problem is that in too many cases, the LLM will only get so far and you will still have to switch to doing actual programming to get the task finished and the worse you get at that last part the more your skillset converges to exactly the type of things an LLM (and therefore everyone else with a keyboard) can reliably do.