GIFT: Games as Informal Training for Generalizable LLMs

2026-01-09

arXiv.org

AI Summary

This work addresses the deficiency of large language models in practical wisdom—such as strategic creativity and social reasoning—typically acquired through informal learning. To bridge this gap, the authors propose leveraging games as informal learning environments, capitalizing on their intrinsic rewards and abstract complexity to cultivate generalizable intelligence. The core innovation lies in a nested training framework that enforces sequential task composition, compelling the model to jointly optimize multiple competencies through an explicit “AND”-style multi-task objective, thereby mitigating task interference. Using GRPO-based reinforcement learning, the model is trained across diverse games including Matrix Games, TicTacToe, and Who's the Spy. Evaluation on multiple capability-oriented benchmarks demonstrates significant improvements in generalization performance, substantiating the efficacy of gamified informal learning for advancing artificial intelligence.

Technology Category

• Game Theory and Economic Paradigms: Adversarial Learning
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Search and Optimization: Learning to Search

Application Category

• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

While Large Language Models (LLMs) have achieved remarkable success in formal learning tasks such as mathematics and code generation, they still struggle with the"practical wisdom"and generalizable intelligence, such as strategic creativity and social reasoning, that characterize human cognition. This gap arises from a lack of informal learning, which thrives on interactive feedback rather than goal-oriented instruction. In this paper, we propose treating Games as a primary environment for LLM informal learning, leveraging their intrinsic reward signals and abstracted complexity to cultivate diverse competencies. To address the performance degradation observed in multi-task learning, we introduce a Nested Training Framework. Unlike naive task mixing optimizing an implicit"OR"objective, our framework employs sequential task composition to enforce an explicit"AND"objective, compelling the model to master multiple abilities simultaneously to achieve maximal rewards. Using GRPO-based reinforcement learning across Matrix Games, TicTacToe, and Who's the Spy games, we demonstrate that integrating game-based informal learning not only prevents task interference but also significantly bolsters the model's generalization across broad ability-oriented benchmarks. The framework and implementation are publicly available.

Problem

• Large Language Models
• informal learning
• generalizable intelligence
• practical wisdom
• social reasoning

Innovation

• informal learning
• nested training framework
• game-based training
• generalizable intelligence
• multi-task reinforcement learning