Unverified50% confidenceFactExact time
Feng et al. (2023) proved that CoT can extend the effective computational complexity of fixed-depth Transformers from the complexity class TC⁰ to the complexity class P.
1
Sources
50%
Confidence
Long-term
Relevance
8/14/2026
First Seen
Sources
Related Claims
UnverifiedTransformer架构的计算复杂度相对序列长度呈二次增长，但高度并行化的特性使其在GPU上的训练效率远超LSTM和GRU架构68% similarUnverifiedMamba、RWKV等线性复杂度架构近年涌现，以应对Transformer的计算瓶颈67% similarUnverified相比RNN/LSTM架构，Transformer支持大规模并行计算67% similarUnverifiedThe core computation of Transformer architecture is matrix multiplication, which aligns with GPU hardware strengths63% similarUnverified深度递归指对单一输入快照的同一状态执行多轮迭代计算，Universal Transformer、DEQ等架构是代表60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/748115API
curl https://kongchang.com/api/v1/knowledge/claims/748115MCP
get_claim(id=748115)