研究揭示LLM智能体记忆盲区：规则被弃用只抄原始日志
Researchers found our current approach to making AI smarter over time has a gian…
Researchers found our current approach to making AI smarter over time has a giant blind spot.
研究人员发现，我们目前让AI随时间变得更聪明的方法存在一个巨大的盲点。
AI is not actually understanding or applying high-level abstract lessons at all.
AI实际上并没有真正理解或应用高层次的抽象教训。
Developers spend massive amounts of time building systems that condense past AI mistakes into neat little rules for the future.
开发者花费大量时间构建系统，将过去的AI错误浓缩成未来简洁的规则。
This paper proves that the AI essentially throws those rules in the trash and only looks at raw historical logs.
这篇论文证明，AI基本上把这些规则扔进了垃圾桶，只查看原始的历史日志。
Modern LLM systems try to get better over time by storing past tasks as either raw step-by-step histories or condensed summary rules. The study tested if these agents actually use their stored memories by secretly swapping the correct tips with random garbage text.
现代LLM系统通过将过去的任务存储为原始的逐步历史或浓缩的摘要规则来尝试随时间改进。研究通过秘密将正确的提示替换为随机垃圾文本来测试这些代理是否真正使用其存储的记忆。
- When the step-by-step histories were messed up, the AI failed hard, proving it heavily relies on copying exact past actions.
- But when researchers completely corrupted the condensed summary rules, the AI kept acting normally and showed zero performance drop.
- 当逐步历史被破坏时，AI表现糟糕，证明它严重依赖复制过去的确切行动。
- 但当研究人员完全破坏浓缩的摘要规则时，AI继续正常运行，性能零下降。
If an AI cannot apply an abstract lesson to a new situation, it is not truly reasoning or learning.
如果AI无法将抽象教训应用到新情况，它就不是真正的推理或学习。
This raises the question if the entire AI industry need to rethink how memory works because right now these agents are just mimicking instead of understanding.
这引发了疑问，整个AI行业是否需要重新思考记忆的工作方式，因为目前这些代理只是在模仿而非理解。
arxiv. org/abs/2601.22436
arxiv. org/abs/2601.22436
"LLM Agents Are Not Always Faithful Self-Evolvers"
"LLM代理并非总是忠实的自我进化者"
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力