![[2602.16488] Learning to Learn from Language Feedback with Social Meta-Learning](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.16488] Learning to Learn from Language Feedback with Social Meta-Learning
Summary
This paper explores a novel approach to enhance large language models (LLMs) by enabling them to learn from language feedback through social meta-learning, improving their adaptability in conversations.
Why It Matters
As LLMs become increasingly integrated into interactive applications, their ability to learn from feedback is crucial for creating more dynamic and effective AI systems. This research addresses significant limitations in current models, paving the way for more human-like interactions and problem-solving capabilities.
Key Takeaways
- Introduces social meta-learning (SML) as a method for LLMs to learn from conversational feedback.
- SML enables models to handle ambiguity better by soliciting necessary information during dialogues.
- Demonstrates that SML-trained models can generalize problem-solving skills across different domains.
- Improves the adaptability of LLMs, making them less prone to premature answers.
- Provides a scalable approach for developing AI systems that effectively learn from interactions.
Computer Science > Computation and Language arXiv:2602.16488 (cs) [Submitted on 18 Feb 2026] Title:Learning to Learn from Language Feedback with Social Meta-Learning Authors:Jonathan Cook, Diego Antognini, Martin Klissarov, Claudiu Musat, Edward Grefenstette View a PDF of the paper titled Learning to Learn from Language Feedback with Social Meta-Learning, by Jonathan Cook and 4 other authors View PDF HTML (experimental) Abstract:Large language models (LLMs) often struggle to learn from corrective feedback within a conversational context. They are rarely proactive in soliciting this feedback, even when faced with ambiguity, which can make their dialogues feel static, one-sided, and lacking the adaptive qualities of human conversation. To address these limitations, we draw inspiration from social meta-learning (SML) in humans - the process of learning how to learn from others. We formulate SML as a finetuning methodology, training LLMs to solicit and learn from language feedback in simulated pedagogical dialogues, where static tasks are converted into interactive social learning problems. SML effectively teaches models to use conversation to solve problems they are unable to solve in a single turn. This capability generalises across domains; SML on math problems produces models that better use feedback to solve coding problems and vice versa. Furthermore, despite being trained only on fully-specified problems, these models are better able to solve underspecified tasks where ...
