Unverified50% confidenceBenchmarkExact time
On long-history tasks, some models can capture a considerable number of correct changes, clearly exceeding the baseline of directly copying the initial state.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
UnverifiedA model's performance must clearly surpass the copy-initial-state baseline to prove that it is genuinely doing state inference rather than static copying.78% similarUnverified与传统可解释性方法（如特征重要性、显著图）不同，机制可解释性追求对模型内部计算的完整逆向70% similarUnverifiedStoring complete tool results in structured state outside the model and sending only brief summaries, exact fields or exact evidence slices usually improves agent quality more than switching from one frontier model to another.69% similarUnverifiedWhenever a model is upgraded, previously accumulated prompting techniques often need to be recalibrated, because the new model's behavioral characteristics, default tendencies, and reasoning approaches may all have changed69% similarUnverifiedExact repetition versus aligned paraphrasing leads to distinctly different contextual usage patterns in the model, depending on the target relationship and the prediction window setting.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1041957API
curl https://kongchang.com/api/v1/knowledge/claims/1041957MCP
get_claim(id=1041957)