[KongchangAI]
Unverified50% confidenceFactExact time

Under the first-order approximation for small weight perturbations, directly adding random biases to the logits can simulate the output change caused by weight perturbation without actually modifying or copying the model weights.

1
Sources
50%
Confidence
Long-term
Relevance
10/30/2026
First Seen

Sources

Related Entities

Related Claims

Cite This Claim

Stable URI
https://kongchang.com/claim/1115745
API
curl https://kongchang.com/api/v1/knowledge/claims/1115745
MCP
get_claim(id=1115745)