# LearnLM: Improving Gemini for Learning

arXiv: https://arxiv.org/abs/2412.16429 · HTML: https://mlsanity.com/paper/2412.16429/

## Key facts

- Cited by 25 papers in the map (data through 14 Oct 2026).
- Builds on 12 of its 25 references that are in the map.
- First posted on arXiv 21 Dec 2024.

## Builds on (most cited first, 12 of 12)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context](https://mlsanity.com/paper/2403.05530/) (2024)
- [Fine-Tuning Language Models from Human Preferences](https://mlsanity.com/paper/1909.08593/) (2019)
- [Instruction-Following Evaluation for Large Language Models](https://mlsanity.com/paper/2311.07911/) (2023)
- [Benchmarking Complex Instruction-Following with Multiple Constraints Composition](https://mlsanity.com/paper/2407.03978/) (2024)
- [InFoBench: Evaluating Instruction Following Ability in Large Language Models](https://mlsanity.com/paper/2401.03601/) (2024)
- [Towards Responsible Development of Generative AI for Education: An Evaluation-Driven Approach](https://mlsanity.com/paper/2407.12687/) (2024)
- [Tutor CoPilot: A Human-AI Approach for Scaling Real-Time Expertise](https://mlsanity.com/paper/2410.03017/) (2024)
- [Towards interactive evaluations for interaction harms in human-AI systems](https://mlsanity.com/paper/2405.10632/) (2024)
- [Assigning AI: Seven Approaches for Students, with Prompts](https://mlsanity.com/paper/2306.10052/) (2023)
- [Human participants in AI research: Ethics and transparency in practice](https://mlsanity.com/paper/2311.01254/) (2023)
- [Teaching and learning in the age of artificial intelligence](https://mlsanity.com/paper/2303.06956/) (2023)

## Built on by (most cited first, 25 of 25)

- [LLMs are Greedy Agents: Effects of RL Fine-tuning on Decision-Making Abilities](https://mlsanity.com/paper/2504.16078/) (2025)
- [EmoBench-M: Benchmarking Emotional Intelligence for Multimodal Large Language Models](https://mlsanity.com/paper/2502.04424/) (2025)
- [WikiBigEdit: Understanding the Limits of Lifelong Knowledge Editing in LLMs](https://mlsanity.com/paper/2503.05683/) (2025)
- [Multi-turn Evaluation of Anthropomorphic Behaviours in Large Language Models](https://mlsanity.com/paper/2502.07077/) (2025)
- [Multimodal Large Language Models Meet Multimodal Emotion Recognition and Reasoning: A Survey](https://mlsanity.com/paper/2509.24322/) (2025)
- [Evaluating Gemini in an arena for learning](https://mlsanity.com/paper/2505.24477/) (2025)
- [Benchmarking the Pedagogical Knowledge of Large Language Models](https://mlsanity.com/paper/2506.18710/) (2025)
- [Pedagogy-driven Evaluation of Generative AI-powered Intelligent Tutoring Systems](https://mlsanity.com/paper/2510.22581/) (2025)
- [FoundationalASSIST: An Educational Dataset for Foundational Knowledge Tracing and Pedagogical Grounding of LLMs](https://mlsanity.com/paper/2602.00070/) (2026)
- [AI tutoring can safely and effectively support students: An exploratory RCT in UK classrooms](https://mlsanity.com/paper/2512.23633/) (2025)
- [Causality Is Key to Understand and Balance Multiple Goals in Trustworthy ML and Foundation Models](https://mlsanity.com/paper/2502.21123/) (2025)
- [A Framework for Situating Innovations, Opportunities, and Challenges in Advancing Vertical Systems with Large AI Models](https://mlsanity.com/paper/2504.02793/) (2025)
- [Large Language Models as Students Who Think Aloud: Overly Coherent, Verbose, and Confident](https://mlsanity.com/paper/2602.01015/) (2026)
- [The Imitation Game for Educational AI](https://mlsanity.com/paper/2502.15127/) (2025)
- [Language Bottleneck Models for Qualitative Knowledge State Modeling](https://mlsanity.com/paper/2506.16982/) (2025)
- [Cultivating Helpful, Personalized, and Creative AI Tutors: A Framework for Pedagogical Alignment using Reinforcement Learning](https://mlsanity.com/paper/2507.20335/) (2025)
- [Hey Chat, Can You Teach Me? Structuring Socratic Dialogue for Human Learning in the Wild](https://mlsanity.com/paper/2606.11744/) (2026)
- [Rethinking LLM-Judged Helpfulness as a Pedagogy Signal: A Pre-Registered Audit Across Tutor Models](https://mlsanity.com/paper/2607.28128/) (2026)
- [Integrating Fine-Grained Audio-Visual Evidence for Robust Multimodal Emotion Reasoning](https://mlsanity.com/paper/2601.18321/) (2026)
- [ISD-Agent-Bench: A Comprehensive Benchmark for Evaluating LLM-based Instructional Design Agents](https://mlsanity.com/paper/2602.10620/) (2026)
- [Sycophancy is an Educational Safety Risk: Why LLM Tutors Need Sycophancy Benchmarks](https://mlsanity.com/paper/2605.14604/) (2026)
- [ConnectED: A Curriculum-Aligned AI System for Vietnamese Instructional Lesson Planning and Student Learning](https://mlsanity.com/paper/2607.28647/) (2026)
- [Auditable Release Control for Pedagogical Leakage in LLM Tutors](https://mlsanity.com/paper/2608.00515/) (2026)
- [Difference-in-Differences on a Censored Rating Scale Can Manufacture an Effect: Evidence from a Pre-Registered LLM-Judge Audit](https://mlsanity.com/paper/2608.27309/) (2026)
- [Sherpa: Teaching LLMs to Teach Adaptively](https://mlsanity.com/paper/2610.08778/) (2026)

## Abstract (arXiv)

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 textitpedagogical 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.