That’s what I’m afraid of for the rest of the development department. They have adopted AI to the core, and barely accept that my part can’t be AI’d for a number of reasons. Once the LLM providers will have to actually charge costs+profits, they’ll lose their whole toolbox, and I doubt they understand what they did to their code base.
Whenever somebody says something like this, there’s the implicit acceptance of the whole “AI is inevitable” trope (spoiler: it isn’t) and the unspoken assumption that AI will be used in the future the same way it’s used now. LLMs were around before the AI craze, and they’ll be around after it. Who knows what they’ll be used for?
I hope I never have to debug the AI crap I’ve seen. My job requires serious thinking about what I do, way deeper than the usual application programming. If I make mistakes, it is not about some app closing, it starts at system crashes and might end with bricked or even burned hardware. Which in turn could become rather expensive. I have to be very disciplined and defensive in my programming, but i also have way better numbers in bugs/loc than others.
The person in the original post said that the frontier models were struggling to make good changes to the AI-ified code base, so what makes you think smaller ones will fare better?
Also, what happens when non-small changes need to be made?
That’s what I’m afraid of for the rest of the development department. They have adopted AI to the core, and barely accept that my part can’t be AI’d for a number of reasons. Once the LLM providers will have to actually charge costs+profits, they’ll lose their whole toolbox, and I doubt they understand what they did to their code base.
deleted by creator
Whenever somebody says something like this, there’s the implicit acceptance of the whole “AI is inevitable” trope (spoiler: it isn’t) and the unspoken assumption that AI will be used in the future the same way it’s used now. LLMs were around before the AI craze, and they’ll be around after it. Who knows what they’ll be used for?
I hope I never have to debug the AI crap I’ve seen. My job requires serious thinking about what I do, way deeper than the usual application programming. If I make mistakes, it is not about some app closing, it starts at system crashes and might end with bricked or even burned hardware. Which in turn could become rather expensive. I have to be very disciplined and defensive in my programming, but i also have way better numbers in bugs/loc than others.
The person in the original post said that the frontier models were struggling to make good changes to the AI-ified code base, so what makes you think smaller ones will fare better?
Also, what happens when non-small changes need to be made?