Expired50% confidenceFactExact time
Anthropic's official best practices explicitly warn that when task descriptions, reports, and code all pile up in the context window, once it is full, the model immediately forgets early instructions and begins to hallucinate.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026(expired)
Sources
Root Causes of Long-Running AI Agent Failures and a Five-Layer Architecture for Control
bilibili老金独立开发6/8/2026
Related Claims
Unverified大模型的幻觉问题目前只能缓解无法根除，知识库约束和精确模式是工程层面最有效的缓解手段69% similarUnverifiedThe larger the framework task, the better models perform; the finer the execution details, the more prone to errors — this is a common problem with large models.68% similarUnverifiedCurrent model instruction-following failures commonly manifest as ignoring system prompt constraints, forgetting early instructions in long conversations, randomly violating formatted output requirements, and deviating from predefined processes in multi-step tasks.68% similarUnverifiedOverly verbose, contradictory, or ambiguous system prompts cause the model's attention to become scattered, resulting in over-thinking or instruction-following failures; streamlining agents.md can improve the determinism of instruction following.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/49996API
curl https://kongchang.com/api/v1/knowledge/claims/49996MCP
get_claim(id=49996)