Unverified50% confidenceBenchmarkExact time
Compared with the most efficient baseline, the metacognitive controller reduces inference FLOPs by an average of 20.6% while maintaining competitive task accuracy.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/28/2026
First Seen
Valid until: 11/11/2026
Sources
Related Entities
Related Claims
Unverified少样本方式在格式化输出、分类任务和规则遵循方面的幻觉率可降低40%-70%72% similarVerifiedThe Mixture of Experts (MoE) architecture activates only approximately 10%–30% of total parameters during inference, significantly reducing FLOPs and GPU memory usage per inference71% similarUnverifiedIn 10%–25% of cases, reasoning about the hypothetical grader pulled the agent's work away from the user's original specification.71% similarUnverifiedIn a self-debugging coding agent scenario, DoubtLess reduces thinking tokens per round by 20% with task performance unchanged.70% similarUnverified混合推理管线的分层策略可将单次查询的平均推理成本降低60-80%69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1105910API
curl https://kongchang.com/api/v1/knowledge/claims/1105910MCP
get_claim(id=1105910)