ParamMem: Augmenting Language Agents with Parametric Reflective Memory

Self-reflection enables language agents to iteratively refine solutions, yet often produces repetitive outputs that limit reasoning performance. Recent studies have attempted to address this limitation through various approaches, among which increasing reflective diversity has shown promise. Our empirical analysis reveals a strong positive correlation between reflective diversity and task success, further motivating the need for diverse reflection signals. We introduce ParamMem, a parametric memory module that encodes cross-sample reflection patterns into model parameters, enabling diverse reflection generation through temperature-controlled sampling. Building on this module, we propose ParamAgent, a reflection-based agent framework that integrates parametric memory with episodic and cross-sample memory. Extensive experiments on code generation, mathematical reasoning, and multi-hop question answering demonstrate consistent improvements over state-of-the-art baselines. Further analysis reveals that ParamMem is sample-efficient, enables weak-to-strong transfer across model scales, and supports self-improvement without reliance on stronger external model, highlighting the potential of ParamMem as an effective component for enhancing language agents.

AI agents often improve their answers by reviewing their own mistakes, a process called self-reflection, but they tend to repeat the same observations and get stuck on the same errors. We show that the variety of these self-reviews matters: the more different angles an assistant considers, the more likely it is to find the right fix. To create that variety, we trained a small, lightweight "memory" module that, once plugged into an existing AI assistant, supplies a fresh perspective at each retry instead of recycled advice. Across programming, math, and multi-step question answering, assistants equipped with our module consistently outperform strong existing methods, work well with only a few hundred training examples, can improve themselves without help from larger AI systems, and even let a small model make a much larger one smarter, offering a simple, broadly useful way to help AI learn from its own mistakes.