Unverified50% confidenceFactExact time
During training, although there is no explicit supervision, the representations developed by the agent's planner can reconstruct the causal structure of the Minkowski metric with high accuracy.
1
Sources
50%
Confidence
Long-term
Relevance
11/3/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedAlignment training optimizes the model's statistical behavior in known scenarios rather than establishing formally verifiable security boundaries, and may fail when faced with rare instruction combinations or role-playing frameworks outside the training distribution.78% similarUnverifiedThe study shows that representation-aware optimization methods can actively shape training trajectories, indicating that the formation geometry is not only measurable but also operable.77% similarUnverifiedTraditional linear probes use maximizing concept classification accuracy as the training objective, without distinguishing whether directions have causal influence.76% similarUnverified任何先优化表示再训练分类器的两阶段方法都需要仔细验证其代理目标与下游任务的对齐程度75% similarUnverifiedA frozen agent refers to an agent whose weights are no longer updated during the inference phase and that relies entirely on capabilities acquired through pre-training or post-training.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1130148API
curl https://kongchang.com/api/v1/knowledge/claims/1130148MCP
get_claim(id=1130148)