<ul data-eligibleForWebStory="false"><li>Researchers at the Allen Institute for Artificial Intelligence have developed a new language model approach named FlexOlmo that enhances privacy of training data.</li><li>FlexOlmo allows multiple companies to jointly train an AI model without sharing their datasets, achieving performance comparable to models trained on a unified dataset.</li><li>Each organization participating in FlexOlmo creates customized models based on an anchor AI model trained on internal data, which are then combined into a single algorithm.</li><li>FlexOlmo decreases the extraction rate of training data, improves model performance by 10.1%, and addresses technical issues by assigning each neural network its own router.</li></ul>

New FlexOlmo language model design removes need for sharing of training data

Discover more