[Qhemlomo]

I'm happy to have a discussion with you if you bring any argument.

Before GPT what would have been your choise of architecture, setup, alogorithm if someone comes to you and says "write a tool/system which can generate code" "what do you mean generate code? How do i control it?" "by writing what you want in natural language" "puh 50 years of development, 100 billion, top tier team of linguists and software engineers perhaps?"

Ask StackOverflow if they think it didn't change anything for them.

[dsr_]

Programming is the reification of decision-making processes. If you don't understand the decision-making process that you want, you get a different one, which at best approximates the one you want but couldn't articulate.

If you do this with COBOL or Python, at least you get consistent operation and errors when you're wrong. If you do this with any LLM, consistency is dropped in favor of obsequiousness.

The base problem is that people aren't equipped naturally to think about all the details of their problems.

[Qhemlomo]

Fine for the initial phase were you have to define the architecture and guardrails etc.

But doesn't matter anymore if you just add things.