{"693089":{"#nid":"693089","#data":{"type":"news","title":"New Neural Framework Could Make LLMs More Empathetic ","body":[{"value":"\u003Cp\u003EAs more people turn to large-language models (LLMs) during personal crises, the companies behind them are engineering these AI-powered models to be more personal, empathetic, and understanding when prompts indicate a user is in emotional distress.\u003C\/p\u003E\u003Cp\u003EThe trend is raising concern about how well prepared ChatGPT, Claude, Gemini, and other popular LLMs are to provide appropriate emotional support.\u003C\/p\u003E\u003Cp\u003ETo address this concern, Georgia Tech researchers have developed a new neural framework that examines how different training sources influence empathy and social reasoning within these AI-powered models.\u003C\/p\u003E\u003Cp\u003E\u201cA lot of people are using LLMs to help process the death of a loved one, or maybe their girlfriend broke up with them, or their boyfriend cheated on them,\u201d said\u0026nbsp;\u003Ca href=\u0022https:\/\/glennmatlin.doctor\/\u0022\u003E\u003Cstrong\u003EGlenn Matlin\u003C\/strong\u003E\u003C\/a\u003E, a Ph.D. student in the School of Interactive Computing.\u003C\/p\u003E\u003Cp\u003EMatlin said the new framework he and a Georgia Tech-led research team have developed audits neural networks with an LLM and estimates the influence documents, transcripts, and other data sources have on a model\u2019s output.\u003C\/p\u003E\u003Cp\u003E\u201cWe\u2019re trying to connect the data to outputs, which is helpful for us in understanding where the information a language model produces comes from,\u201d he said.\u003C\/p\u003E\u003Cp\u003EMatlin is the lead author of a new paper examining model sourcing for social reasoning and how models mimic human behavior in conversations. The\u0026nbsp;\u003Ca href=\u0022https:\/\/hcai-lab-gt.github.io\/capabilibara\/\u0022\u003Epaper\u003C\/a\u003E will be presented this week at the\u0026nbsp;\u003Ca href=\u0022https:\/\/colm.cc\/\u0022\u003EConference on Language Modeling\u003C\/a\u003E in San Francisco.\u003C\/p\u003E\u003Cp\u003EMatlin said the findings suggest that an objective tone in its output indicates that a model places greater importance on textbooks or similar neutral training sources. An empathetic tone suggests the model values sources with dialogue and conversational examples more.\u003C\/p\u003E\u003Cp\u003E\u201cWe find that models tend to learn better from interpersonal dialogue when it comes to social reasoning,\u201d he said. \u201cModels that are good at social reasoning will rely less on textbooks about human emotions and more on FAQs or customer service dialogue.\u003C\/p\u003E\u003Cp\u003E\u201cIt can read all it wants from texts that explain anger, but that wouldn\u2019t help it determine whether someone is angry.\u201d\u003C\/p\u003E\u003Cp\u003E\u003Ca href=\u0022https:\/\/eilab.gatech.edu\/mark-riedl.html\u0022\u003E\u003Cstrong\u003EProfessor Mark Riedl\u003C\/strong\u003E\u003C\/a\u003E, Matlin\u2019s advisor, said that knowing which sources models value most could enable engineers to alter their behavior.\u003C\/p\u003E\u003Cp\u003E\u201cThe question is, in this vast ocean of documents in this neural network, which ones become important in certain situations, and which ones don\u2019t,\u201d Riedl said.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u201cThis leads to the potential for interventions in how we train the model. If we want more empathy, we might look to see where it learns empathy and give it more of those documents, fewer of them, or other documents.\u201d\u003C\/p\u003E\u003Cp\u003ERiedl said the framework can audit an LLM\u2019s sourcing for advice on any topic, from financial to medical.\u003C\/p\u003E\u003Cp\u003E\u201cWe want to help people make more informed decisions about whether they should use AI for a particular task,\u201d he said.\u003C\/p\u003E\u003Cp\u003E\u201cYou might make a different decision about whether to trust the AI\u2019s recommendation if you knew how a skill was learned and which documents influenced that decision. The first step is knowing how it learns, where it learns from, and what documents tend to lead to certain behaviors.\u201d\u003C\/p\u003E\u003Cp\u003EThe findings that Matlin, Riedl, and their collaborators have made so far came through a\u0026nbsp;\u003Ca href=\u0022https:\/\/allenai.org\/blog\/olmo-capability-tracing\u0022\u003Epartnership with the Allen AI Institute\u003C\/a\u003E (Ai2), which allowed them to audit their open-sourced LLM, Olmo 3.\u003C\/p\u003E\u003Cp\u003EWhile not as large as ChatGPT or Claude, Olmo 3\u2019s training data comprises over a billion documents.\u003C\/p\u003E\u003Cp\u003EMatilin said it\u2019s impossible to know why ChatGPT or Claude behave in certain ways because their parent companies do not disclose their training data. However, Matlin\u2019s study may help researchers to make better guesses about what\u2019s happening behind the scenes.\u003C\/p\u003E\u003Cp\u003EHe added that the study is the first of its kind and scale on an academic level.\u003C\/p\u003E\u003Cp\u003E\u201cThis is the first one that\u2019s ever been done for an open data model of that type,\u201d he said. \u201cWe\u2019re providing a view into neural networks that is hard to come by without being in the corporate setting.\u201d\u003C\/p\u003E","summary":"","format":"limited_html"}],"field_subtitle":"","field_summary":[{"value":"\u003Cp\u003EGeorgia Tech researchers have developed a new neural framework that examines how different training sources influence empathy and social reasoning within large language models.\u003C\/p\u003E","format":"limited_html"}],"field_summary_sentence":[{"value":"Georgia Tech researchers have developed a new neural framework that examines how different training sources influence empathy and social reasoning within large language models."}],"uid":"36530","created_gmt":"2026-10-05 19:48:24","changed_gmt":"2026-10-05 19:49:20","author":"Nathan Deen","boilerplate_text":"","field_publication":"","field_article_url":"","location":"Atlanta, GA","dateline":{"date":"2026-10-05T00:00:00-04:00","iso_date":"2026-10-05T00:00:00-04:00","tz":"America\/New_York"},"extras":[],"hg_media":{"681361":{"id":"681361","type":"image","title":"ACM-AI-2026_86A6750.jpg","body":null,"created":"1791229727","gmt_created":"2026-10-05 19:48:47","changed":"1791229727","gmt_changed":"2026-10-05 19:48:47","alt":"Mark Riedl","file":{"fid":"265741","name":"ACM-AI-2026_86A6750.jpg","image_path":"\/sites\/default\/files\/2026\/10\/05\/ACM-AI-2026_86A6750.jpg","image_full_path":"http:\/\/hg.gatech.edu\/\/sites\/default\/files\/2026\/10\/05\/ACM-AI-2026_86A6750.jpg","mime":"image\/jpeg","size":123514,"path_740":"http:\/\/hg.gatech.edu\/sites\/default\/files\/styles\/740xx_scale\/public\/2026\/10\/05\/ACM-AI-2026_86A6750.jpg?itok=pWipndaV"}}},"media_ids":["681361"],"groups":[{"id":"47223","name":"College of Computing"},{"id":"1188","name":"Research Horizons"},{"id":"50876","name":"School of Interactive Computing"}],"categories":[{"id":"194606","name":"Artificial Intelligence"},{"id":"153","name":"Computer Science\/Information Technology and Security"}],"keywords":[],"core_research_areas":[{"id":"193655","name":"Artificial Intelligence at Georgia Tech"},{"id":"39501","name":"People and Technology"}],"news_room_topics":[],"event_categories":[],"invited_audience":[],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[],"email":[],"slides":[],"orientation":[],"userdata":""}}}