Hacker News Reader

Comment by fassssst | Hacker News Reader

Right, hopefully no one is sprinkling eval(prompt) into their codebase.

I had an idea about that the other day!

What if you could write something like:

  @implement_this
  def prime_sieve(n: int) -> list[int]:
    pass

And the decorator reads the function name and optional docstring, runs it through an LLM and replaces the function with one implementing the desired behavior (hopefully correctly). I remember there was something like this for StackOverflow.

https://github.com/retrage/gpt-macro

I made that as a joke in javascript 8 years ago. Not using LLM, but using javascript proxies so that if a function doesn't exist it tries to implement it runtime based on the name of the function using a sql like grammar. I really hope to not see something like that in real use, heh.

https://github.com/Matsemann/Declaraoids

Maybe I should make a LLM version of this.

The funny part is, Spring Data JPA has a quite serious take on this, in the form of query methods [0]. You create a repository interface with certain method names, and it dynamically creates an implementation that queries the columns of the underlying table, according to the name of the method.

[0] https://docs.spring.io/spring-data/jpa/reference/jpa/query-m...

I've never tried it myself, but Prefect does have something like this with their Marvin AI library for Python.

https://github.com/PrefectHQ/marvin?tab=readme-ov-file#-buil...

I'm building magentic https://github.com/jackmpcollins/magentic which has basically this syntax, though it queries the LLM to generate the answer rather than writing + running code.

  from magentic import prompt
  from pydantic import BaseModel
  
  class Superhero(BaseModel):
      name: str
      age: int
      power: str
      enemies: list[str]
  
  @prompt("Create a Superhero named {name}.")
  def create_superhero(name: str) -> Superhero: ...

I do have plans to also solve the case you're talking about of generating code once and executing that each time.

Not gonna lie... llama.cpp... LlamaIndex... Ollama... kinda is eval(prompt) a lot of the time!

Of course AI data pipelines are a totally different conversation than code solutions.

Reply on news.ycombinator.com