LearnLM: Improving Gemini for Learning
Abstract
Today's generative AI systems are tuned to present information by default, rather than engage users in service of learning as a human tutor would. To address the wide range of potential education use cases for these systems, we reframe the challenge of injecting pedagogical behavior as one of \textit{pedagogical instruction following}, where training and evaluation examples include system-level instructions describing the specific pedagogy attributes present or desired in subsequent model turns. This framing avoids committing our models to any particular definition of pedagogy, and instead allows teachers or developers to specify desired model behavior. It also clears a path to improving Gemini models for learning -- by enabling the addition of our pedagogical data to post-training mixtures -- alongside their rapidly expanding set of capabilities. Both represent important changes from our initial tech report. We show how training with pedagogical instruction following produces a LearnLM model (available on Google AI Studio) that experts substantially prefer across a diverse set of learning scenarios, with average preference strengths of +31\% over GPT-4o, +11\% over Claude 3.5 Sonnet, and +13\% over the Gemini 1.5 Pro model on which LearnLM was based.
Cite
@article{arxiv.2412.16429,
title = {LearnLM: Improving Gemini for Learning},
author = {LearnLM Team and Abhinit Modi and Aditya Srikanth Veerubhotla and Aliya Rysbek and Andrea Huber and Brett Wiltshire and Brian Veprek and Daniel Gillick and Daniel Kasenberg and Derek Ahmed and Irina Jurenka and James Cohan and Jennifer She and Julia Wilkowski and Kaiz Alarakyia and Kevin R. McKee and Lisa Wang and Markus Kunesch and Mike Schaekermann and Miruna Pîslar and Nikhil Joshi and Parsa Mahmoudieh and Paul Jhun and Sara Wiltberger and Shakir Mohamed and Shashank Agarwal and Shubham Milind Phal and Sun Jae Lee and Theofilos Strinopoulos and Wei-Jen Ko and Amy Wang and Ankit Anand and Avishkar Bhoopchand and Dan Wild and Divya Pandya and Filip Bar and Garth Graham and Holger Winnemoeller and Mahvish Nagda and Prateek Kolhar and Renee Schneider and Shaojian Zhu and Stephanie Chan and Steve Yadlowsky and Viknesh Sounderajah and Yannis Assael},
journal= {arXiv preprint arXiv:2412.16429},
year = {2025}
}