Unverified50% confidenceFactExact time
In the glassy state, attention inference is unstable, and the system may fall into a large number of local minima, leading to unreliable inference results.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedOverly 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.73% similarUnverified模型在大海捞针测试中对中间位置信息的检索能力明显弱于首尾位置，这被称为注意力沉降现象73% similarUnverifiedAn embedding variance threshold that is too low may misclassify normal focused thinking as a loop.72% similarUnverified在长对话中早期的错误尝试和调试过程会持续占据上下文窗口，造成上下文污染并干扰后续推理准确性72% similarUnverifiedThe base rate fallacy points out that when targets are extremely rare, even a highly accurate system may produce a large proportion of false alarms among its alerts.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1023888API
curl https://kongchang.com/api/v1/knowledge/claims/1023888MCP
get_claim(id=1023888)