2 ms·

So, assembly?

• goosejuice 4mo ago
  That would be less efficient
  • pjmlp 4mo ago
    There are already plenty of PLDI and SIGPLAN talks on LLM => Assembly, with some guardrails for determinism, instead of having a "classical" language as translation layer.
    • goosejuice 4mo ago
      Sorry are we referring to the efficiency of generating programs (i.e. optimizing LLM software development) or the efficiency of the output (i.e. optimizing software through the generation of assembly)?
      • pjmlp 4mo ago
        We are referring to not having to use 3GL languages as intermediate step for LLMs to generate native code. "Programming Language Design and Implementation in the Era of Machine Learning" https://www.youtube.com/watch?v=Fc3cW0nqAQ0 https://www.youtube.com/watch?v=Fc3cW0nqAQ0
        • goosejuice 4mo ago
          I'm not seeing strong claims here on improved token efficiency for general use cases, but it's absolutely not my area of expertise. Completely understandable why this would be valuable research though. Will be interesting to see how it evolves.