Agreed. History shows concentration in the hands of a few usually doesn't go well, not because the masses are powerful or smarter than the few but there is a limit you can manage power from the top. AI will be another step in the popularization of information and we'll get both Wikipedia and TikTok of AI in coming years.
No one here understands 'sikisinca' as 'sikişince', because as you can see the final vowel is different. There are cases for these to be mixed though, sikildim May be "I'm fscked" but it's usually understood as "sıkıldım", "I'm bored."
What I have in mind as an analogy is house cleaning or laundry. In the past these would consume so much time and when washing machines appeared, they probably didn't wash as the "best washer" but they let a large percentage of population to spend less time with laundry and people began to see having clean clothes as a standard.
As you said the current state of LLMs is not dependable but we'll learn how to deal with their poor quality and come up with solutions. The overall secondary effect may be increased standards for the knowledge work everywhere. No one would bother to read TOS for a SaaS, now you can ask for a summary from LLM, or ask it to write a privacy notice to your blog.
I believe LLMs will create more jobs than it eliminated by raising standards in various fields, including software development.
We will have to get to 100% test coverage and document everything and add more bells and whistles to UI etc. The day to day activity may change but there will always be developers.
I deleted "Senior ML/MLOps Engineer" from LinkedIn bio and wrote "Data Carpenter, Software Plumber, Container Architect, Model Trainer, Robot Guide" yesterday :))
Recently I tried to translate a few 100-200 line scripts from zsh to nushell with Claude 3.5 Sonnet and it sucked. This is now my go to experiment for new LLMs, translating code between two programming languages must be easier than translating natural language to a programming language, yet we don't see any such results, even for popular languages.
This is my experience as well. I wrote Telegram bots in Python and Go for myself, they had weird bugs difficult to debug. Then I wrote it in Rust and it works. I don't have to go back every three weeks to reset the server. Would I write it in C++? No way.
"I couldn't horseshoe a hard systems language to replace a web framework for a CRUD app written with soft language, so I had to go back"
Rails is very good until you need to go off rails. Even in that case rewrites must be partial and not to aim to fully replace the original. Most of the time rewrites are a waste of time. This article could be true if you s/Rust/Ruby and tell the story in the opposite direction.
If you're working with text and you have some regularities in your workflow, you can gain some productivity points from current technology. In time most industries will have these regularities, similar to workflow improvements after the introduction of computer.
But I also doubt if it will have that much effect. In general we have diminishing returns from these kind of technology, telegraph's effect on the world was larger than the computer, computers were more valuable than WWW and search engines have smaller effect than WWW. AI may be big, but could it be bigger than search or e-commerce? I doubt it.