I was thinking about how Advent Of Code and how coding has changed. I think it could be fun to participate in a challenge oriented website where the solution had to be 'programmed' using a prompt using a local small LLM. The server would then check the correctness of the solution and somehow verify that the solution was generated with the small LLM.
I think it could be a good way to also teach and explore prompting styles with a specific objective in mind.
Does anything like this exist or does it sound interesting for someone to build?
There are some hurdles to overcome:
- what LLM would everyone use that is small enough to run on a webgpu (for example)
- how would the server verify the code was generated by the LLM and not tampered with (if that's even possible)
