zlacker

[parent] [thread] 0 comments
1. timhig+(OP)[view] [source] 2025-05-23 17:27:36
This paper seems to focus on highly algorithmic/puzzle-like problems, which are not the typical application domain of LLMs, using a <500M parameter model. So my hunch is "reasoning" works much better for math, coding, factual recall, and writing tasks that most LLMs actually deal with.
[go to top]