Ask HN: Languages in the Age of AI

1 points by HeyLaughingBoy 14 days ago | 3 comments

Most of my professional work is in embedded systems, typically using C++. I've been doing more of my hobby work in Python (MicroPython/CircuitPython) and was thinking of investigating Rust.

But then it occurred to me that if we're using AI more and more to generate code, does it really matter what language is used? If an agent is writing most of my code from prompts, I just need to understand the language well enough to verify that the reviewing agent is doing a good enough job and maybe tweak it a bit when I have to. If we can trust the AI to avoid the common problems, then what difference does the language make?

dabinat 14 days ago |

I’d argue that it still makes a difference because the language can determine what kinds of issues are possible to occur, as well as the quality and availability of training data from that language. Rust is a good language for AI because it gives the agent immediate feedback if something is wrong.

But it may be the case that AI-specific languages and higher abstractions evolve and in future we may just be managing a series of prompts that generate code on the fly at compile time.

naivehacker 13 days ago | |

> But it may be the case that AI-specific languages and higher abstractions evolve and in future we may just be managing a series of prompts that generate code on the fly at compile time.

I am glad to this idea but I think it won't be natural languages - it may be some kind of Immediate Representation - full of semantic definite tokens.

naivehacker 13 days ago |

AI writes Python most and best. Rust is a choice, because its compiler has more constraints on code semantics - the borrow checker. AI struggles against Rust compiler, and it deserves. Other languages, such as C++, I don't think AI will have good support on them - if a reasoning chain for writing code is hard to human, then it will be also hard to AI. AI will cause the same amount of bugs.