我给我的交易代理添加了记忆，结果一切变得更糟

An experiment with an LLM trading agent revealed that its own memory led to self-destruction over two days. This issue was not a mere bug but a structural flaw in how similarity-based memory systems interact with the internal beliefs of LLMs. The surprisingly simple solution involved making the agent remember its failures, even when the retrieval system resisted this. This led to the development of an open-source mechanism designed to prevent such feedback loops in the future.