Saphra will be an assistant professor in Computing & Data Sciences<\/strong>), has worked for over a decade in the growing field of interpretability, in which researchers poke around inside language models to uncover the mechanisms that make them work. While many of her fellow interpretability researchers draw inspiration from neuroscience, Saphra favors a different analogy. Interpretability, in her view, should take a cue from evolutionary biology.<\/p>\n\u201cThere\u2019s this very famous quote by [the geneticist Theodosius] Dobzhansky: \u2018Nothing makes sense in biology except in the light of evolution,\u2019\u201d she said. \u201cNothing makes sense in AI except in the light of stochastic gradient descent,\u201d a classic algorithm that plays a central role in the training process through which large language models learn to generate coherent text.<\/p>\n
Language models are based on neural networks, mathematical structures that process data using connections between artificial \u201cneurons.\u201d The strength of each connection is random at first, but during the training process the connections get tweaked as the model repeatedly attempts to predict the next word in sentences from a vast text dataset. Somehow, through trillions of tiny tweaks, the model develops internal structures that enable it to \u201cgeneralize,\u201d or respond fluently to unfamiliar inputs.<\/p>\n
Most interpretability research focuses on understanding these structures in language models after the training process. Saphra is a prominent champion of an alternative approach that focuses on the training process itself. Just as biologists must understand an organism\u2019s evolutionary history to fully understand the organism, she argues, interpretability researchers should pay more attention to what happens during training. \u201cIf you don\u2019t understand the origins of the model, then you don\u2019t understand why anything works,\u201d she said.<\/p>\n
Read the Full Article<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"Take a look at the recent feature of incoming CDS Assistant Professor Naomi Saphra, and her work with LLM behavior to optimize and improve how they are trained and utilized.<\/p>\n","protected":false},"author":18839,"featured_media":18305,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[251,5,35,274,244],"tags":[377,314,884],"_links":{"self":[{"href":"https:\/\/www.bu.edu\/cds-faculty\/wp-json\/wp\/v2\/posts\/18304"}],"collection":[{"href":"https:\/\/www.bu.edu\/cds-faculty\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.bu.edu\/cds-faculty\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.bu.edu\/cds-faculty\/wp-json\/wp\/v2\/users\/18839"}],"replies":[{"embeddable":true,"href":"https:\/\/www.bu.edu\/cds-faculty\/wp-json\/wp\/v2\/comments?post=18304"}],"version-history":[{"count":1,"href":"https:\/\/www.bu.edu\/cds-faculty\/wp-json\/wp\/v2\/posts\/18304\/revisions"}],"predecessor-version":[{"id":18306,"href":"https:\/\/www.bu.edu\/cds-faculty\/wp-json\/wp\/v2\/posts\/18304\/revisions\/18306"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.bu.edu\/cds-faculty\/wp-json\/wp\/v2\/media\/18305"}],"wp:attachment":[{"href":"https:\/\/www.bu.edu\/cds-faculty\/wp-json\/wp\/v2\/media?parent=18304"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.bu.edu\/cds-faculty\/wp-json\/wp\/v2\/categories?post=18304"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.bu.edu\/cds-faculty\/wp-json\/wp\/v2\/tags?post=18304"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
![\"Photo](\"\/cds-faculty\/files\/2025\/09\/Naomi-Saphra-Quanta-Magazine-Post.jpg\")
<\/p>\n