Unverified50% confidenceBenchmarkExact time
Replay experiment: fixing the winning lr/wd schedule, without using PBT, and running on fresh initializations, 55/60 reached 95% accuracy; whereas with fixed hyperparameters at the geometric mean of that schedule, 0/60 succeeded.
1
Sources
50%
Confidence
Long-term
Relevance
10/26/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedReAct's errors amplify in a chain: if the accuracy per step is p, the overall success rate after n steps is approximately p to the power of n; with a per-step accuracy of 95%, the overall success rate drops to approximately 59.9% after 10 steps and only 35.8% after 20 steps.76% similarUnverifiedLLM每次调用可建模为伯努利试验，N次调用通过次数服从二项分布B(N, p)，通过率95%置信区间约为p̂ ± 1.96×√(p̂(1-p̂)/N)70% similarUnverifiedDistilBERT retains approximately 97% of the original model's performance on the GLUE benchmark, with inference speed improved by approximately 60% and model size reduced by 40%.69% similarUnverifiedIn the benchmark test, the heuristic peak rule achieved 93.3% overall accuracy, 100.0% impact recall, and 38.4% cadence false positives, requiring manual threshold tuning.69% similarUnverified探索性能上界实验发现现有验证粒度并非最优，理想情况下计算量（FLOPs）可减少75倍，同时准确率提升1.1%69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1099240API
curl https://kongchang.com/api/v1/knowledge/claims/1099240MCP
get_claim(id=1099240)