Unverified50% confidenceBenchmarkExact time
Compared with direct continuous-time sampling, SpecCTMC reduces the number of sequential model evaluations by 1.73 to 4.38 times while maintaining the reference sampling distribution.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/6/2026
First Seen
Valid until: 2/4/2027
Sources
Related Entities
Related Claims
UnverifiedOn the WildChat dataset, the cascaded scheme reduced the number of judge calls by 97.1% relative to direct LLM judgment.74% similarUnverified一项针对GPT-4的研究显示，对MMLU题目进行语义等价改写后，模型准确率在部分子集上下降超过10个百分点74% similarUnverifiedApplying the synthetic control method to bookings per thousand people rather than raw booking volumes allows the model to track the treatment group more closely in the pre-treatment period, with RMSE dropping to 4.4 percentage points.73% similarUnverifiedExperiments on an eight-dimensional Gaussian mixture dataset show that this scheduling strategy requires only 303 measurements per sample to achieve generation quality comparable to a high-measurement reference of 1.6×10^6 measurements, a reduction of about 5300 times in the number of measurements.71% similarUnverifiedAfter adding expert-guided instructions, the models' search time and context usage increase by about 1.6 to 2.7 times, and forecast error decreases accordingly.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1142343API
curl https://kongchang.com/api/v1/knowledge/claims/1142343MCP
get_claim(id=1142343)