Large language models (LLMs) have unveiled remarkable reasoning capabilitiesby exploiting chain-of-thought (CoT) prompting, which generates intermediatereasoning chains to serve as the rationale for deriving the answer. However,current CoT methods either simply employ general prompts such as Let's thinkstep by step, or heavily rely on handcrafted task-specific demonstrations toattain preferable performances, thereby engendering an inescapable gap betweenperformance and generalization. To bridge this gap, we propose Meta-CoT, ageneralizable CoT prompting method in mixed-task scenarios where the type ofinput questions is unknown. Meta-CoT firstly categorizes the scenario based onthe input question and subsequently constructs diverse demonstrations from thecorresponding data pool in an automatic pattern. Meta-CoT simultaneously enjoysremarkable performances on ten public benchmark reasoning tasks and superiorgeneralization capabilities. Notably, Meta-CoT achieves the state-of-the-artresult on SVAMP (93.7%) without any additional program-aided methods. Ourfurther experiments on five out-of-distribution datasets verify the stabilityand generality of Meta-CoT.
URL
Affiliations
Abstract
Translation (by gpt-3.5-turbo)
Summary (by gpt-3.5-turbo)