Fast-Slow Training uses context optimization as fast weights alongside parameter updates as slow weights to achieve up to 3x better sample efficiency, higher performance, and less catastrophic forgetting than standard RL in continual LLM learning.
Fernando Hernandez-Garcia, Qingfeng Lan, Parash Rahman, A
6 Pith papers cite this work, alongside 117 external citations. Polarity classification is still indexing.
6
Pith papers citing it
117
external citations · OpenAlex
citation-role summary
background 2
citation-polarity summary
years
2026 6verdicts
UNVERDICTED 6roles
background 2polarities
background 2representative citing papers
Aurora is a leverage-aware spectral optimizer that enforces uniform row norms in matrix updates while preserving Muon's polar geometry, outperforming Muon and achieving SOTA among spectral methods on modded-nanoGPT.
Task diversity along map, object, and hierarchy axes produces local transfer across shifts in a new continual RL benchmark but fails to sustain learning as the number of shifts grows.
Sharpness-aware pretraining and related flat-minima interventions reduce catastrophic forgetting by up to 80% after post-training across 20M-150M models and by 31-40% at 1B scale.
Loss smoothing interpolates between source and target objectives during adaptation and improves performance across supervised shifts, vision fine-tuning, RL, and LM tasks.
The paper reframes agentic safety as an epistemic property defined by teachability—the capacity to preserve future corrective leverage—rather than a behavioral property of the current policy.
citing papers explorer
-
Learning, Fast and Slow: Towards LLMs That Adapt Continually
Fast-Slow Training uses context optimization as fast weights alongside parameter updates as slow weights to achieve up to 3x better sample efficiency, higher performance, and less catastrophic forgetting than standard RL in continual LLM learning.
-
Aurora: A Leverage-Aware Spectral Optimizer
Aurora is a leverage-aware spectral optimizer that enforces uniform row norms in matrix updates while preserving Muon's polar geometry, outperforming Muon and achieving SOTA among spectral methods on modded-nanoGPT.
-
Task diversity produces systematic transfer but inhibits continual reinforcement learning
Task diversity along map, object, and hierarchy axes produces local transfer across shifts in a new continual RL benchmark but fails to sustain learning as the number of shifts grows.
-
Sharpness-Aware Pretraining Mitigates Catastrophic Forgetting
Sharpness-aware pretraining and related flat-minima interventions reduce catastrophic forgetting by up to 80% after post-training across 20M-150M models and by 31-40% at 1B scale.
-
Loss Smoothing for Stable Adaptation Under Distribution Shift
Loss smoothing interpolates between source and target objectives during adaptation and improves performance across supervised shifts, vision fine-tuning, RL, and LM tasks.
-
Agentic Safety is an Epistemic Property, Not a Behavioral One
The paper reframes agentic safety as an epistemic property defined by teachability—the capacity to preserve future corrective leverage—rather than a behavioral property of the current policy.