Unverified50% confidenceFactExact time
In very long conversations, information located in the middle is severely downweighted by the model's attention mechanism, while content at the beginning and end of the conversation is assigned the highest weight—the so-called lost-in-the-middle problem.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe memory of most agents is limited to the context window of a single conversation, with state reset to zero after the task ends, making it impossible to accumulate experience across sessions.72% similarUnverifiedIn extremely long contexts, the model's attention gradually dilutes, which is the technical root cause of AI gradually forgetting the requirements stated at the very beginning.72% similarUnverified当系统提示词因多任务而变得复杂时，不同任务的指令之间会产生语义冲突，稀释注意力机制对关键信息的聚焦能力71% similarUnverifiedDemoting chat to a floating popup breaks the user's mental model, which is a classic navigation hierarchy misalignment problem.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/892971API
curl https://kongchang.com/api/v1/knowledge/claims/892971MCP
get_claim(id=892971)