Unverified50% confidenceOpinionExact time
Compared with SFT, RL training is more unstable, more sensitive to hyperparameters, and more compute-hungry.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedFor 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 instability74% similarUnverifiedEric Bigelow believes that many capabilities already exist in pre-trained base models, and that subsequent SFT, RL, and agent fine-tuning are more about sharpening the distribution than endowing entirely new capabilities.73% similarUnverifiedThe essence of choosing a simulation platform is a trade-off between RL training efficiency and physical fidelity72% similarUnverifiedRLHF训练中会加入KL散度惩罚项，防止模型偏离SFT阶段学到的基础行为太远71% similarUnverifiedRLHF和DPO等对齐技术的效果高度依赖预训练基础质量，预训练阶段未习得的能力几乎无法通过后续对齐补救71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1070722API
curl https://kongchang.com/api/v1/knowledge/claims/1070722MCP
get_claim(id=1070722)