In recent times, Large Language Models (LLMs) have evolved and transformed Natural Language Processing with their few-shot prompting techniques.\u00a0 These models have extended their usability in almost every domain, ranging from Machine translation, Natural Language Understanding, Text completion, sentiment analysis, speech recognition, and so on.\u00a0 With the few-shot prompting approach, LLMs are provided with a few examples of a particular task, along with some natural language instructions, and using these; they are able to adapt and learn how to perform the task properly.\u00a0 The tasks requiring iterative steps and constraint propagation come with many limitations when using these prompting techniques, to overcome which a new approach has been introduced.<\/p>\n
A team of researchers at Microsoft Research, Redmond, USA, recently introduced a new method called Reprompting, which addresses all the limitations accompanying prompting techniques.\u00a0 This approach automatically searches for some useful and effective chain-of-thought (CoT) prompts.\u00a0 Chain-of-thought prompting helps improve the reasoning ability of large language models and helps them perform complex reasoning tasks.\u00a0 For this, a few chains of thought demonstrations are provided as exemplars during prompting.\u00a0 Reprompting finds CoT prompts very efficiently without any human involvement.\u00a0<\/p>\n
The researchers have used an iterative sampling approach known as Gibbs sampling in the Reprompting algorithm.\u00a0 It frames the problem as sampling from a joint distribution of CoT recipes.\u00a0 Since the distribution is difficult to characterize directly, Gibbs Sampling has been used as an approximation method.\u00a0 This sampling method helps determine the best instructions by trying different ones and deciding which works best.<\/p>\n
The Reproompting algorithm begins with a sampling of initial CoT recipes with the help of zero-shot prompting, where no prompt information is provided.\u00a0 Zero-shot prompting enables an LLM to generate task responses without prior training.\u00a0 The algorithm then iteratively samples new recipes using previously sampled solutions as parent prompts, and these new recipes are used to solve other training problems, aiming to find a set of prompts that share similar CoT prompts.\u00a0<\/p>\n
The algorithm has been evaluated on the five Big-Bench Hard (BBH) tasks that require multi-step reasoning.\u00a0 BBH focuses on tasks that are believed to be beyond the abilities and potentials of the current language models.\u00a0 ChatGPT and InstructGPT have been used as LLMs for the evaluation of the algorithm.\u00a0 Upon evaluation, Reprompting has proved to perform better than the zero-shot, few-shot, and human-written CoT prompting techniques.\u00a0<\/p>\n
Reprompting also showed significant potential in model combination by using different LLMs for initializing and sampling new recipes.\u00a0 It can help in the transfer of knowledge from a stronger model to a weaker model, thus resulting in a noticeably better performance shown by the weaker model.\u00a0 Reprompting performed better than the human-written CoT prompting on BBH tasks by up to 17 points.\u00a0 The researchers have mentioned that the CoT recipes that work fine on one model may not work well on another, highlighting the need for optimizing CoT for each model to have some fairer comparisons.<\/p>\n
To sum up, the Reprompting algorithm is a great automated approach for finding effective CoT prompts for LLMs without human intervention.\u00a0 It is a valuable approach to addressing the limitations of existing methods and achieving superior performance on tasks requiring multi-step reasoning.<\/p>\n
Check out the\u00a0Paper<\/strong>.<\/a> Don\u2019t forget to join\u00a0our 21k+ ML SubReddit<\/strong>,\u00a0Discord Channel<\/strong>,<\/a>\u00a0and\u00a0Email Newsletter<\/a><\/strong>, where we share the latest AI research news, cool AI projects, and more. If you have any questions regarding the above article or if we missed anything, feel free to email us at\u00a0Asif@marktechpost.com<\/strong><\/p>\n Check Out 100\u2019s AI Tools in AI Tools Club<\/strong><\/a><\/p>\n The post Microsoft Researchers Introduce Reprompting: An Iterative Sampling Algorithm that Searches for the Chain-of-Thought (CoT) Recipes for a Given Task without Human Intervention<\/a> appeared first on MarkTechPost<\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":" In recent times, Large Language Models (LLMs) have evolved and transformed Natural Language Processing with their few-shot prompting techniques.\u00a0 These models have extended their usability in almost every domain, ranging from Machine translation, Natural Language Understanding, Text completion, sentiment analysis, speech recognition, and so on.\u00a0 With the few-shot prompting approach, LLMs are provided with a […]<\/p>\n","protected":false},"author":0,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-22165","post","type-post","status-publish","format-standard","hentry","category-news"],"_links":{"self":[{"href":"https:\/\/healthmedicinet.com\/business\/wp-json\/wp\/v2\/posts\/22165","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/healthmedicinet.com\/business\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/healthmedicinet.com\/business\/wp-json\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/healthmedicinet.com\/business\/wp-json\/wp\/v2\/comments?post=22165"}],"version-history":[{"count":0,"href":"https:\/\/healthmedicinet.com\/business\/wp-json\/wp\/v2\/posts\/22165\/revisions"}],"wp:attachment":[{"href":"https:\/\/healthmedicinet.com\/business\/wp-json\/wp\/v2\/media?parent=22165"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/healthmedicinet.com\/business\/wp-json\/wp\/v2\/categories?post=22165"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/healthmedicinet.com\/business\/wp-json\/wp\/v2\/tags?post=22165"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}