Unverified50% confidenceFactExact time
A recent paper specifically studies the gradient descent (GD) dynamics of adversarial training (AT) with large step sizes, revealing three key differences from standard training (ST).
1
Sources
50%
Confidence
Long-term
Relevance
11/6/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe change in gradient routing may affect training stability, especially when the loss surface is steeper in the early stage of training, and it is recommended to start with small-scale experiments for validation.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% similarUnverifiedThe study compared standard methods, including supervised fine-tuning (SFT), GRPO, and gradient ascent unlearning.69% similarUnverified不同模型在抗诱导能力上的差异很大程度上源于后训练阶段（post-training）的策略差异68% similarUnverifiedThe double descent phenomenon and the grokking phenomenon show that training dynamics are more complex than traditional learning theory predicts.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1142244API
curl https://kongchang.com/api/v1/knowledge/claims/1142244MCP
get_claim(id=1142244)