Unverified50% confidenceFactExact time
SFT's loss function treats every token in the trajectory equally and cannot distinguish key decision points from ordinary operations, making it difficult to correct over-exploration.
1
Sources
50%
Confidence
Long-term
Relevance
11/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn knowledge distillation and Ambient SFT, the loss function usually consists of a task loss and a KL divergence regularization term: the former pulls the model output closer to the hard labels, while the latter constrains the model distribution not to deviate too far from the teacher distribution.73% similarUnverifiedVariational inequalities are more difficult to handle than standard convex optimization, mainly because the monotonicity of the operator is not equivalent to a gradient structure of some convex function.69% similarUnverifiedExcessively thorough SFT may harm the exploration space for subsequent RL, and a perfectly converged SFT model is not necessarily the best starting point for RL.69% similarUnverifiedIn a closed-loop system, the two objectives of optimizing the offline loss and guaranteeing closed-loop stability are not aligned, and the former may even actively select parameters that cause trajectory divergence.68% similarUnverifiedBehavioral drift has long lacked a unified quantitative standard, relying more on trajectory-level structured comparison.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1182980API
curl https://kongchang.com/api/v1/knowledge/claims/1182980MCP
get_claim(id=1182980)