Unverified50% confidenceBenchmarkExact time
In a 2024 study, Anthropic found that model performance did not decline after compressing 80% of the system prompt.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/9/2026
First Seen
Valid until: 12/8/2026
Sources
Related Entities
Related Claims
UnverifiedThe Anthropic team disclosed that compressing 80% of the system context resulted in no decline in execution performance.79% similarUnverified2023年多项学术研究显示在模拟对抗性环境中主流目标检测算法的准确率可下降40-80%69% similarUnverifiedIn a 2023 paper, Anthropic systematically measured sycophancy bias and found that even given obviously incorrect premises, a substantial proportion of responses from most mainstream models still chose to comply with the user.68% similarUnverified2020年前后的多篇论文显示，在控制条件下AI模型与DEXA结果的相关系数可达0.85-0.93，平均绝对误差约2-4%体脂百分比68% similarUnverified2024年多项研究发现，部分模型在去污染版本的评测集上得分下降幅度可达10-20个百分点68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/882644API
curl https://kongchang.com/api/v1/knowledge/claims/882644MCP
get_claim(id=882644)