Unverified50% confidenceFactExact time
In autoregressive models, gradient updates at different response positions share the underlying representation and parameters, so supervision of a certain deep token can transfer to other positions; this phenomenon is called cross-position supervision transfer.
1
Sources
50%
Confidence
Long-term
Relevance
11/17/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe two mechanisms of prospective valuation and provenance-gated credit assignment can be transferred to other agent scenarios that require state passing across tasks.67% similarUnverifiedPrompting techniques are transferable across different models and different platforms.67% similarUnverified权重共享使卷积操作等价于对输入信号做互相关运算,并引入平移等变性这一归纳偏置66% similarUnverifiedTo assess whether a model has regressed, one should fix the input entry point, fix the prompt, and fix the task for a horizontal comparison, rather than relying on subjective impressions.65% similarUnverifiedThe intra-group mean is a scalar shared across the entire trajectory and cannot distinguish the contribution of each step within the trajectory, corresponding to the problem that the same feedback signal is applied to every step.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1179758API
curl https://kongchang.com/api/v1/knowledge/claims/1179758MCP
get_claim(id=1179758)