it is quite terrifying that people think translating from formal constraints in math to informal english is either accurate or deterministic.

And they ask the LLM to not look at requirements but who is guaranteeing it won’t just hallucinate them out of thin air later when messy prompts inevitably trigger a butterfly effect?

Safe you all the click. Author claims ai is getting better at writing code and we should think about what we want not how we want it.

Friendly reminder LLMs are not good a writing software. They can only put pieces together that may make sense.

Want proof? Ask it to write some valid C++26 without raw pointers, using modules and reflection.