Unverified50% confidenceFactExact time
Taking a Transformer as an example, the output of each neuron or attention head can be viewed as a direction in a high-dimensional space, and changing the projection strength along that direction systematically affects downstream computation.
1
Sources
50%
Confidence
Long-term
Relevance
10/28/2026
First Seen
Sources
Related Entities
Related Claims
Unverified各向异性现象的根源在于Transformer的自注意力机制和残差连接结构,使高频词的梯度信号主导嵌入空间整体方向78% similarVerifiedThe self-attention mechanism of the Transformer can capture dependencies between any two positions in a sequence, and regardless of how far apart the tokens are, the computational cost of calculating attention weights is the same78% similarVerifiedBecause the Transformer self-attention mechanism exhibits positional bias, earlier input content has a stronger anchoring effect on subsequent generation, which is the underlying reason for the context anchoring effect of system prompts76% similarUnverifiedFor Transformers, common interpretability methods include analyzing attention weights, tracing neuron activations, or using probes.75% similarUnverifiedFrom PCA dimensionality reduction to the attention mechanism in Transformers, matrix operations and vector space transformations are fundamental.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1108614API
curl https://kongchang.com/api/v1/knowledge/claims/1108614MCP
get_claim(id=1108614)