CoT prompts use a step-by-step explanation to guide a large language model to develop an answer. CoT prompts have been shown to significantly increase productivity in activities that require extensive reasoning. The self-consistency (SC) technique further improves accuracy by sampling multiple thought chains and returning the majority output.
Efficiency gains are the result of SC, but the method is flawed. The first is that it is impossible to reach a consensus when there are many conceivable outcomes because each chain of reasoning could end with a different result. Second, ignoring the thought process that led to the result can cause you to miss important details.
In their paper “Multi-chain Reasoning,” researchers from Tel Aviv University, the Allen Institute for Artificial Intelligence, and Bar Ilan University present a method called MCR, in which they instruct a large language model (LLM) to meta-reason. through various chains of reasoning. and generate a conclusive answer and explanation. The sampled chains of reasoning are not used for their predictions (as in SC), but rather to collect data from multiple chains. While both approaches rely on drawing from a pool of possible chains of reasoning, SC gives the most commonly reached answer for those chains: “No” (gray box, bottom right). By contrast, MCR combines the intermediate results of each string (blue boxes, top left) into a single context that is then delivered to a meta-reasoner model along with the original query. The meta-reasoner is a distinct LLM that is asked to meta-reason through several different lines of reasoning before arriving at a conclusive solution and justification.
The core of MCR consists of three parts. The chain of reasoning is generated by combining a decomposition model and a retriever. After combining these strings, a multi-string context is created and fed to the meta-reasoner.
The team tests MCR on numerous difficult multi-hop QC data sets in an open domain scenario. They classify problems as implicit or explicit. They use SC and versions of Self-Ask and CoT with recovery as benchmarks for comparisons with MCR. Using the same number of chains of reasoning, the results reveal that MCR consistently outperforms all other baselines. They assess the value of MCR by carefully scoring and measuring the quality of the explanations it generates. According to the findings, MCR can produce well-reasoned explanations for more than 82% of situations.
review the Research work and GitHub link. Don’t forget to join our 20k+ ML SubReddit, discord channel, and electronic newsletter, where we share the latest AI research news, exciting AI projects, and more. If you have any questions about the article above or if we missed anything, feel free to email us at [email protected]
🚀 Check out 100 AI tools at AI Tools Club
Tanushree Shenwai is a consulting intern at MarktechPost. She is currently pursuing her B.Tech at the Indian Institute of Technology (IIT), Bhubaneswar. She is a data science enthusiast and has a strong interest in the scope of application of artificial intelligence in various fields. She is passionate about exploring new advances in technology and its real life application.
