https://www.arxiv.org/abs/2404.02255
https://old.reddit.com/r/MachineLearning/comments/1c0grto/r_a_simple_society_of_language_models_solves/
* modularizes decomposition, solution, verification into 3 different language models
* outperforms best baselines by 8.1% on MATH, 7.71% on JEEBench, 9.7% on MedQA problems
#LLM #ML #reasoning
