[gametorch]

I appreciate the reply.

What is everyone working on that takes more than five minutes to think about?

For me, the work is insurmountable and infinite, while coming up with the solution is never too difficult. I'm not saying this to be cocky. I mean this:

In 99.9999999999% of the problems I encounter in software engineering, someone smarter than me has already written the battle tested solution that I should be using. Redis. Nginx. Postgres. etc. Or it's a paradigm like depth first search or breadth first search. Or just use a hash set. Sometimes it's a little crazier like Bloom filters but whatever.

Are you like constantly implementing new data structures and algorithms that only exist in research papers or in your head?

Once you've been engineering for 5 or 10 years, you've seen almost everything there is to see. Most of the solutions should be cached in your brains at that point. And the work just amounts to tedious, unimportant implementation details.

Maybe I'm forgetting that people still get bogged down in polymorphism and all that object oriented nonsense. If you just use flat structs, there's nothing too complicated that could possibly happen.

I worked in HFT, for what it's worth, and that should be considered very intense non-CRUD "true" engineering. That, I agree, LLMs might have a little more trouble with. But it's still nothing insane.

Software engineering is extremely formulaic. That's why it's so easy to statistically model it with LLMs.

[nextlevelwizard]

I write embedded software in C++ for industrial applications. We have a lot of proprietary protocols and custom hardware. We have some initiatives to train LLMs with our protocols/products/documentation, but I have not been impressed with the results. Same goes with our end-to-end testing framework. I guess it isn't so popular so the results vary a lot.

I have been doing this for 8 year and while yes I have seen a lot you can't just copy-paste solutions due to flash, memory, and performance constraints.

Again maybe this is a skill issue and maybe I will be replaced with an LLM, but so far they seem more like cool toys. I have used LLMs to write AddOns for World of Warcraft since my Lua knowledge is mostly writing Wireshark plugins for our protocols and for that it has been nice, but it is nothing someone who actually works with Lua or with WoW API couldn't produce faster or just as fast, because I have to describe what I want and then try and see if the API the LLM provides exists and if it works as the LLM assumed it would.

[gametorch]

Again, I appreciate the reply. I think my view on LLMs is skewed towards the positive because I've only been building CRUD apps, command line tools, and games with them. I apologize if I came off as incendiary or offensive.