Ask HN: Are we going to see new AI-friendly programming languages?
3 comments
Nah, everything will just migrate to Python :P
Seriously though, there is a chicken-and-egg problem where current LLMs favor more popular instruction and programming languages. I like the idea of human friendly constructed language being translated a hyper optimized middle language (MLIR? A C or Rust subset?), but generating the training dataset seems like a difficult problem.
Maybe a smaller neural net/interface could translate the "LLM programing language" to tokens?
Seriously though, there is a chicken-and-egg problem where current LLMs favor more popular instruction and programming languages. I like the idea of human friendly constructed language being translated a hyper optimized middle language (MLIR? A C or Rust subset?), but generating the training dataset seems like a difficult problem.
Maybe a smaller neural net/interface could translate the "LLM programing language" to tokens?
Something like a low-code system that smooths away the details? Yes.
You are on to something.
This may be the hundred billion dollar industry of the future (forget quantum coding, the qubit is a dead end.)
I admit I have devised a practical linguistic system with long term dreams of similar goals. I wonder how many closet nerds have done the same.
Features of this (conceptualized) linguistic system
- differentiation between an abstract assertion and a specific instance (and identity)
- complete articulation of subjective scoping (all systems are interoperations of subjective scopes)
- distinction between a terminal and non-terminal “fact”
- expansive handling of “truth” and “truthiness”
I call this the “Objectified Systemic Language” and it’s really about assertions and relationships and less on linear logic.
While logic and rules and constraints do come into things, this linguistic approach would postulate assertions and find resolve among potentials.
For instance, one might define morality as “do not destructively interfere with that which is inculpable”. Along with more intrinsic definitions of destructive interference, and culpability a competent learning system should work out that any circumstance in which one subjective destructively interferes with another subjective (without a lawful right for instance) is BAD.
I find intelligence to be more like chemistry, and we deceive ourselves as to the rational sequential order of our own intellectual processes.
This may be the hundred billion dollar industry of the future (forget quantum coding, the qubit is a dead end.)
I admit I have devised a practical linguistic system with long term dreams of similar goals. I wonder how many closet nerds have done the same.
Features of this (conceptualized) linguistic system
- differentiation between an abstract assertion and a specific instance (and identity)
- complete articulation of subjective scoping (all systems are interoperations of subjective scopes)
- distinction between a terminal and non-terminal “fact”
- expansive handling of “truth” and “truthiness”
I call this the “Objectified Systemic Language” and it’s really about assertions and relationships and less on linear logic.
While logic and rules and constraints do come into things, this linguistic approach would postulate assertions and find resolve among potentials.
For instance, one might define morality as “do not destructively interfere with that which is inculpable”. Along with more intrinsic definitions of destructive interference, and culpability a competent learning system should work out that any circumstance in which one subjective destructively interferes with another subjective (without a lawful right for instance) is BAD.
I find intelligence to be more like chemistry, and we deceive ourselves as to the rational sequential order of our own intellectual processes.
It could be argued that programming languages have traditionally been an interface between the human brain the the computer, so as AI becomes better at natural language processing, are we going to see more simplified programming languages formed around the process of instructing an AI system?