Verified65% confidenceFactTime unknown
During SFT training, the model only computes loss on the assistant portion (loss masking technique), meaning it only learns how to answer, not how to ask questions.
3
Sources
65%
Confidence
Long-term
Relevance
6/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSupervised 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."80% similarUnverifiedFor classification tasks with discrete and closed label spaces, the training signal from SFT is already clear enough, and introducing reinforcement learning would instead increase the difficulty of hyperparameter tuning and training instability75% similarUnverifiedIn the post-training phase, Motif adopts a "failure-driven SFT" strategy: identifying model weaknesses through benchmark testing and internal scenarios, then generating targeted datasets for training.75% 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.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/39837API
curl https://kongchang.com/api/v1/knowledge/claims/39837MCP
get_claim(id=39837)