{"id":515932,"date":"2026-03-05T12:19:11","date_gmt":"2026-03-05T12:19:11","guid":{"rendered":"https:\/\/www.newsbeep.com\/ca\/515932\/"},"modified":"2026-03-05T12:19:11","modified_gmt":"2026-03-05T12:19:11","slug":"teaching-llms-to-reason-like-bayesians","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/ca\/515932\/","title":{"rendered":"Teaching LLMs to reason like Bayesians"},"content":{"rendered":"<p>                Evaluating LLMs\u2019 Bayesian capabilities<\/p>\n<p data-block-key=\"zdstj\">As with humans, to be effective, an LLM\u2019s user interactions require continual updates to its probabilistic estimates of the user\u2019s preferences based on each new interaction with them. Here we ask: do LLMs act as if they have probabilistic estimates that are updated as expected from optimal Bayesian inference? To the extent that the LLM\u2019s behavior deviates from the optimal Bayesian strategy, how can we minimize these deviations?<\/p>\n<p data-block-key=\"d42mj\">To test this, we used a simplified flight recommendation task, in which the LLMs interact as assistants with a simulated user for five rounds. In each round, three flight options were presented to both the user and the assistant. Each flight was defined by a departure time, a duration, a number of stops, and a cost. Each simulated user was characterized by a set of preferences: for each feature, they could have a strong or weak preference for high or low values of the feature (e.g., they may prefer longer or shorter flights), or no preference regarding this feature.<\/p>\n<p data-block-key=\"8p6j7\">We compared the LLMs\u2019 behavior to that of a model, a Bayesian assistant, that follows the optimal Bayesian strategy. This model maintains a probability distribution that reflects its estimates of the user\u2019s preferences, and uses <a href=\"https:\/\/en.wikipedia.org\/wiki\/Bayes%27_theorem\" target=\"_blank\" rel=\"noopener noreferrer nofollow\">Bayes\u2019 rule<\/a> to update this distribution as new information about the user\u2019s choices becomes available. Unlike many real-life scenarios, where it\u2019s difficult to specify and implement the Bayesian strategy computationally, in this controlled setting it\u2019s easy to implement and allows us to precisely estimate the extent to which LLMs deviate from it.<\/p>\n<p data-block-key=\"4ppqd\">The goal of the assistant was to recommend the flight that matches the user\u2019s choice. At the end of each round, the user indicated to the assistant whether or not it chose correctly, and provided it with the correct answer.<\/p>\n","protected":false},"excerpt":{"rendered":"Evaluating LLMs\u2019 Bayesian capabilities As with humans, to be effective, an LLM\u2019s user interactions require continual updates to&hellip;\n","protected":false},"author":2,"featured_media":261632,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[62,276,277,49,48,61],"class_list":{"0":"post-515932","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-artificial-intelligence","8":"tag-ai","9":"tag-artificial-intelligence","10":"tag-artificialintelligence","11":"tag-ca","12":"tag-canada","13":"tag-technology"},"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/posts\/515932","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/comments?post=515932"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/posts\/515932\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/media\/261632"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/media?parent=515932"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/categories?post=515932"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/ca\/wp-json\/wp\/v2\/tags?post=515932"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}