Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note3 min read

AutoMem: Automated Learning of Memory as a Cognitive Skill

Source

arXiv

arxiv.org/abs/2607.01224v1

Read the source

Source

Plain-Language Brief

Researchers have developed a framework called AutoMem to improve the performance of large language models (LLMs) in long-horizon games by treating memory management as a learnable skill. They let the model decide how to manage its memory and optimize it over two axes: the structure that supports memory (prompts, file schemas, action vocabulary), and the proficiency of the model exercising it. By automating both axes, they were able to improve the base agent's performance by up to 4x, making the model competitive with frontier systems like Claude Opus 4.5 and Gemini 3.1 Pro Thinking.

What the Document Says

  • Memory expertise is treated as a learnable skill in LLMs.
  • File-system operations are promoted to first-class memory actions.
  • The model decides how to manage its memory, with improvements along two axes: the structure supporting memory and the proficiency of the model.
  • AutoMem automates both axes by having a strong LLM review complete agent trajectories and iteratively revise the memory structure.
  • The agent's good memory decisions are used as training signals to sharpen the model's memory proficiency directly.
  • AutoMem was tested on three procedurally generated long-horizon games, with optimizing memory alone improving the base agent's performance 2x-4x.
  • This improvement made the open-weight model competitive with Claude Opus 4.5 and Gemini 3.1 Pro Thinking.

Key Concepts Mentioned

  • Large Language Models (LLMs)
  • Memory management
  • Metamemory
  • Long-horizon games
  • Automated learning
  • AutoMem framework

Claims Made by the Document

  • Memory expertise is a learnable skill.
  • Optimizing memory alone can improve the performance of LLMs in long-horizon tasks significantly.

Evidence Used

  • Improved base agent's performance 2x-4x in long-horizon games by optimizing memory alone.
  • Brought a 32B open-weight model competitive with frontier systems like Claude Opus 4.5 and Gemini 3.1 Pro Thinking.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Memory expertise is a learned skill: knowing what to encode, when to retrieve, and how to organize knowledge--a capacity known in cognitive science as metamemory. We bring this perspective to LLMs by treating memory management as a trainable skill. We promote file-system operations to first-class memory actions alongside task actions, letting the model itself decide how to manage its memory. This memory skill improves along two axes: the structure that supports it (prompts, file schemas, action vocabulary), and the proficiency of the model exercising it. Both axes resist manual optimization: episodes in long-horizon tasks run for thousands of steps, and a single memory mistake can hide long before it surfaces, making human review of full trajectories impractical. We introduce AutoMem, a framework that automates both axes. In the first loop, a strong LLM reviews complete agent trajectories and iteratively revises the memory structure that shapes how the agent interacts with its memory files. In the second loop, the agent's own good memory decisions are identified from many episodes and used as training signal to sharpen the model's memory proficiency directly. Across three procedurally generated long-horizon games (Crafter, MiniHack, and NetHack), optimizing memory alone--without modifying the model's task-action behavior--improved the base agent's performance ~2x-4x, bringing a 32B open-weight model competitive with frontier systems such as Claude Opus 4.5 and Gemini 3.1 Pro Thinking. Our results show that memory management is an independently learnable skill, and a high-leverage objective yielding large gains on long-horizon tasks.