Unverified50% confidenceBenchmarkExact time
The researchers used Grid-Bench to evaluate six LLMs and found that as the operation history grows longer, the accuracy of predicting the complete state drops sharply.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
Unverified研究表明LLM在处理过长、过复杂的指令集时整体遵循率会系统性下降，即指令遗忘现象69% similarVerifiedGradient descent iteratively adjusts weights in the opposite direction of the loss function's gradient, causing the prediction error to converge progressively.67% similarUnverifiedThe researchers transfer the population-level conclusions to finite corpora by controlling the gradient error so that it remains manageable relative to the shrinking progress directions along the scale of each parameter67% similarUnverifiedLLMs still have clear shortcomings in predicting multi-step changes in discrete worlds, and their ability to track spatial operations and infer states gradually decays over long chains.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1041952API
curl https://kongchang.com/api/v1/knowledge/claims/1041952MCP
get_claim(id=1041952)