Unverified50% confidenceFactExact time
WISD constructs weakened models through two approaches, low-bit quantization and magnitude pruning; in the method these two operations are used to generate training signals rather than for deployment optimization.
1
Sources
50%
Confidence
Long-term
Relevance
10/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWISD adopts on-policy self-distillation, where the training signal comes from comparing parallel trajectories generated in real time by the original model and the weakened model on the same unlabeled prompt, with the comparison occurring in policy space rather than answer space.79% similarUnverifiedThe refined annotations are ultimately used to fine-tune a lightweight model; at deployment, the low-cost lightweight model performs large-scale extraction, with the LLM intervening only once during the training data construction phase.70% similarUnverifiedA contemporaneous study by O'Brien et al. found that when mid-training was used to teach the model that misaligned behavior is acceptable within a specified context (marked with the invented token <quarantine_token>), and SFT/RL training data conforming to that format was used, misalignment generalization could be reduced, but not beyond the effect of inoculation prompting.69% similarUnverifiedOn infilling tasks, Diffusion-LM outperforms continuous-relaxation-based methods such as COLD and DELOREAN on automatic evaluation metrics, and reaches the level of autoregressive models trained from scratch specifically for this purpose without requiring targeted training.69% similarUnverified权重归一化技术将权重向量分解为方向与模长两个独立分量分别优化，可显著改善训练稳定性68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1114231API
curl https://kongchang.com/api/v1/knowledge/claims/1114231MCP
get_claim(id=1114231)