Programming languages are dead.
As interesting as the debate can be about moving from C to Rust, or how the hell is Python the official language of AI, what I think the question should be is whether any of that matters now that we can instruct the computer to write code, any type of code.
The future programmer, reviewing and testing code written by itself. (image generated by Grok)
I started to ask various LLMs to write assembly code and to optimize it to specific hardware, and they seem to cope really well, to the point where I'm thinking about the idea that we should just kill all languages and stick with assembly, now that the majority of hurdles are removed:
Assembly is hard to learn, and time consuming to write - That's ok because we're not typing it all by hand.
Assembly is a low level language, difficult to read, without abstractions - True but that was only valid when people were writing and most importantly, debugging that code. We now live in the most abstract of concepts, human language!
Assembly implementations can be quite fast and the benefits can be even greater if we can further tune it by hand to be even faster.
Also, why say that Assembly is great when most of the training has been done on a multitude of languages? Why can't the binary code of all software be used to train it, or in other words, since the output of all different languages is often assembly code, and if that is fed in to training data, can't it be even better than using GitHub repos?
Then, if AGI is something that happens, then why won't the machine invent a new and better programming language that all AI's will use to write code, which in turn, will end up as assembly opcodes anyway.
Why bother with computer languages, and the so called stacks, if the new programmer is a manager first, a quality control technician second, and a production sysadmin in total.
In sum, I question if the time where we invent new programming languages has indeed come to an end?
