generalize
The paper mainly describes a way to use thinking chain to improve the reasoning ability of LLM model, and through experiments, it proves the significant effect of thinking chain in tasks such as arithmetic, common sense and symbols. The sota effect on GSM8K can be achieved only through the 540B size PaLM model and 8 thought chain samples.
Specific work
This paper mainly explores the reasoning ability of the LLM model. The reasoning ability of the LLM model can be enhanced through reinforcement learning and ICL, but the effect of reinforcement learning and ICL improvement is limited and the cost is relatively high. Based on the above status quo and inspiration, this paper proposes a method based on thinking chains to enhance the reasoning ability of LLM, and has achieved good results.
And the thinking chain has the above attractive characteristics.