Search papers, labs, and topics across Lattice.
This paper investigates the challenges of continual experience internalization in large language models (LLMs), revealing that multi-iteration learning leads to a capability collapse rather than improvement. The authors identify that principle-level experience is more effective than instance-level experience, and that step-wise experience injection outperforms global injection by better aligning with decision states. By employing off-policy context-distillation on high-quality teacher trajectories, they provide a more stable training signal, offering a robust framework for developing self-evolving LLMs.
Multi-iteration experience learning in LLMs can lead to capability collapse, but strategic adjustments in experience granularity and injection patterns can stabilize and enhance performance.
Experience internalization converts contextual experience from past interactions into reusable parametric capability, offering a promising path toward continual learning in large language models (LLMs). While prior work has predominantly focused on single-iteration transfer, we discover that under multi-iteration experience learning, existing methods suffer from a progressive capability collapse rather than compounding improvement. We systematically examine this failure through three vital dimensions of experience internalization: (1) Experience Granularity: We find that principle-level experience is more durable than instance-level experience, as it effectively abstracts transferable strategies away from trajectory-specific details. (2) Experience Injection Pattern: Our analysis reveals that step-wise injection significantly outperforms global injection by aligning experience with intermediate decision states, a property that is critical for long-horizon tool use. (3) Internalization Regime: We demonstrate that off-policy context-distillation on high-quality teacher trajectories provides a substantially more stable training signal than on-policy context-distillation, which is inherently limited by local corrections on student-induced flawed states. Together, these insights yield a simple yet robust recipe for stable and sustainable experience internalization, providing concrete guidance for engineering self-evolving and continually learning LLMs.