Unverified50% confidenceFactExact time
级联架构中每个模块独立训练，上游模块的错误会不可逆地传递到下游
1
Sources
50%
Confidence
Long-term
Relevance
8/7/2026
First Seen
Sources
Gemini 3.5实时字幕实测：电竞解说场景下的语音转录表现
Related Claims
UnverifiedNaïve post-training quantization causes individual weight errors to accumulate and amplify across multiple layers—a phenomenon known as compounding error—and at 2-bit or 1-bit precision the model structure collapses73% similarUnverified级联失败是指单次工具调用错误会污染后续上下文，导致模型在错误的中间状态上继续推理，最终使整个任务偏轨72% similarUnverified仅用成功轨迹训练的模型往往存在分布外泛化的严重缺陷，遇到工具返回错误时便丧失纠偏能力70% similarUnverifiedBehavior Cloning (BC) suffers from the covariate shift problem: during training the model only sees the state distribution of expert trajectories, and at deployment small errors can drive the robot into states not covered by the training set, leading to cascading failures.70% similarUnverifiedThe effectiveness of traditional ensemble learning depends on the errors among base learners being mutually independent; when error patterns are highly correlated, the ensemble benefit shrinks significantly.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/705482API
curl https://kongchang.com/api/v1/knowledge/claims/705482MCP
get_claim(id=705482)