_version_ 1866912547709386752
author LearnLM Team
Modi, Abhinit
Veerubhotla, Aditya Srikanth
Rysbek, Aliya
Huber, Andrea
Wiltshire, Brett
Veprek, Brian
Gillick, Daniel
Kasenberg, Daniel
Ahmed, Derek
Jurenka, Irina
Cohan, James
She, Jennifer
Wilkowski, Julia
Alarakyia, Kaiz
McKee, Kevin R.
Wang, Lisa
Kunesch, Markus
Schaekermann, Mike
Pîslar, Miruna
Joshi, Nikhil
Mahmoudieh, Parsa
Jhun, Paul
Wiltberger, Sara
Mohamed, Shakir
Agarwal, Shashank
Phal, Shubham Milind
Lee, Sun Jae
Strinopoulos, Theofilos
Ko, Wei-Jen
Wang, Amy
Anand, Ankit
Bhoopchand, Avishkar
Wild, Dan
Pandya, Divya
Bar, Filip
Graham, Garth
Winnemoeller, Holger
Nagda, Mahvish
Kolhar, Prateek
Schneider, Renee
Zhu, Shaojian
Chan, Stephanie
Yadlowsky, Steve
Sounderajah, Viknesh
Assael, Yannis
author_facet LearnLM Team
Modi, Abhinit
Veerubhotla, Aditya Srikanth
Rysbek, Aliya
Huber, Andrea
Wiltshire, Brett
Veprek, Brian
Gillick, Daniel
Kasenberg, Daniel
Ahmed, Derek
Jurenka, Irina
Cohan, James
She, Jennifer
Wilkowski, Julia
Alarakyia, Kaiz
McKee, Kevin R.
Wang, Lisa
Kunesch, Markus
Schaekermann, Mike
Pîslar, Miruna
Joshi, Nikhil
Mahmoudieh, Parsa
Jhun, Paul
Wiltberger, Sara
Mohamed, Shakir
Agarwal, Shashank
Phal, Shubham Milind
Lee, Sun Jae
Strinopoulos, Theofilos
Ko, Wei-Jen
Wang, Amy
Anand, Ankit
Bhoopchand, Avishkar
Wild, Dan
Pandya, Divya
Bar, Filip
Graham, Garth
Winnemoeller, Holger
Nagda, Mahvish
Kolhar, Prateek
Schneider, Renee
Zhu, Shaojian
Chan, Stephanie
Yadlowsky, Steve
Sounderajah, Viknesh
Assael, Yannis
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2412_16429
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle LearnLM: Improving Gemini for Learning
LearnLM Team
Modi, Abhinit
Veerubhotla, Aditya Srikanth
Rysbek, Aliya
Huber, Andrea
Wiltshire, Brett
Veprek, Brian
Gillick, Daniel
Kasenberg, Daniel
Ahmed, Derek
Jurenka, Irina
Cohan, James
She, Jennifer
Wilkowski, Julia
Alarakyia, Kaiz
McKee, Kevin R.
Wang, Lisa
Kunesch, Markus
Schaekermann, Mike
Pîslar, Miruna
Joshi, Nikhil
Mahmoudieh, Parsa
Jhun, Paul
Wiltberger, Sara
Mohamed, Shakir
Agarwal, Shashank
Phal, Shubham Milind
Lee, Sun Jae
Strinopoulos, Theofilos
Ko, Wei-Jen
Wang, Amy
Anand, Ankit
Bhoopchand, Avishkar
Wild, Dan
Pandya, Divya
Bar, Filip
Graham, Garth
Winnemoeller, Holger
Nagda, Mahvish
Kolhar, Prateek
Schneider, Renee
Zhu, Shaojian
Chan, Stephanie
Yadlowsky, Steve
Sounderajah, Viknesh
Assael, Yannis
Computers and Society
Artificial Intelligence
Machine Learning
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.
title LearnLM: Improving Gemini for Learning
topic Computers and Society
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2412.16429