Unverified50% confidenceFactExact time
Current mainstream stateless dialogue models lose all memory after each conversation ends and are unable to initiate actions autonomously.
1
Sources
50%
Confidence
Long-term
Relevance
10/6/2026
First Seen
Sources
Related Entities
Related Claims
Unverified提示词工程的核心缺陷在于'无状态性'，每次对话都是全新开始，上下文和角色设定需要重复注入77% similarUnverifiedIn 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.75% similarVerifiedThe 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.73% similarUnverifiedFirst-generation Prompt-stage Agents have no memory, reason only based on the current input, and the context disappears after the conversation ends.73% similarUnverifiedState and conversation should be separated; constraints require a persistent, lossless, and verifiable medium, whereas conversation is inherently transient and lossy.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/980211API
curl https://kongchang.com/api/v1/knowledge/claims/980211MCP
get_claim(id=980211)