The topic is a bit provocative, but I am interested to hear opinion of people who understand how programing languages and compilers work. According to some resources dedicated to AI discussion, many software developers are actively using AI assistants for their work. Some claim that many devs do from 30% to 60% of their work using AI. Suppose, this trend continues, and over 90% of coding will be done by AI in the future. In this case, what will be the point in generating intermediate code in a mix of programming languages (C, python, julia etc) and maintaining thousands of libraries, instead of giving AI technical specs in human language (English, Chinese etc) and asking it to generate machine code directly? Even if some intermediate code is necessary, say to keep it platform agnostic, why should we use pre-ai languages, instead of some brand new language, which is optimized for AI from the start? I it possible to use AI to create such a language?
I’ve done a little work building chatbots, and that little experience trying to get LLMs to return something useful highlighted the verbosity required to unambiguously describe an algorithm (or anything, really) in natural language. It makes one realize the reasons for (relatively) concise, unambiguous notations such as programming languages or mathematics. It’s a situation similar to legal contract language–so much verbiage to define everything without loopholes.
If the future of programming amounts to prompt engineering, I’m glad that I’m nearing retirement.
Did calculators make the need for math disappear?
Nobody is saying math will become obsolete. The way we use to transform the math into machine instructions may change significantly. Or may be not.