elvis@omarsar0DAIR.AI 的 Elvis Saravia 推荐了一项研究：RLM 的智能体框架（harness）可作为组合泛化器，使 Transformer 无需额外泛化能力即可迁移任务。实验表明，RLM 仅在短任务上训练，就能完全泛化到长度 8-32 倍的未见任务，且跨领域任务（如数学与写作）共享分解策略时也呈现相同效果。
Highly recommended.
I've often claimed there's huge alpha in building agent harnesses.
Turns out harnesses are compositional generalizers. The RLM harness is an instance of this. This could lead to interesting and efficient new ways to scale generalization in models.
elvis@omarsar0 · XDAIR.AI 的 Elvis Saravia 推荐了一项研究：RLM 的智能体框架（harness）可作为组合泛化器，使 Transformer 无需额外泛化能力即可迁移任务。实验表明，RLM 仅在短任务上训练，就能完全泛化到长度 8-32 倍的未见任务，且跨领域任务（如数学与写作）共享分解策略时也呈现相同效果。
Highly recommended.
I've often claimed there's huge alpha in building agent harnesses.
Turns out harnesses are compositional generalizers. The RLM harness is an instance of this. This could lead to interesting and efficient new ways to scale generalization in models.