3
Pitted GPT-4 against a smaller local model for a code review task and the little guy won
I had GPT-4 rewrite a whole Python function last week but when I ran it through Llama 3.2 locally it caught two edge cases the big model totally missed, so has anyone else found smaller models outperforming the giant ones on specific boring tasks?
2 comments
Log in to join the discussion
Log In2 Comments
the_sandra4d ago
Honestly that makes total sense, the smaller models are way more focused on specific patterns while the big ones try to be too smart and overcomplicate things. Ngl I've noticed Llama 3.2 just seems to catch those little gotchas in code that the big models breeze past.
6
spencer_perry774d ago
Maybe the smaller models are better at code review because they weren't trained to be creative writers too, just pure logic and pattern matching. The big ones have too much fluff in their heads from all that poetry and marketing copy they've digested.
5