Unverified50% confidenceFactExact time
The design of SSD keeps training rollouts close to the student's own test-time behavior, avoiding the problems caused by distribution shift in off-policy methods.
1
Sources
50%
Confidence
Long-term
Relevance
11/10/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThrough targeted teacher intervention, SSD avoids the low-signal state that readily arises in full-policy rollouts, because the teacher provides signals only at positions where knowledge injection is genuinely needed.76% similarUnverifiedPre-training and SFT provide only positive supervision, whereas DPO and adjacent RL techniques provide the leverage of negative supervision, which can reduce the probability of erroneous sequences.67% similarUnverifiedSupervised fine-tuning (SFT) optimizes imitation loss rather than a behavioral safety objective, so it readily exhibits the phenomenon of "working within the training distribution but failing out of distribution."65% similarUnverified维护多样化对手池、保留不同课程阶段的模型快照并在采样时兼顾各阶段,有助于防止训练陷入狭窄的策略循环65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1155428API
curl https://kongchang.com/api/v1/knowledge/claims/1155428MCP
get_claim(id=1155428)