Unverified50% confidenceFactExact time
Pre-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.
1
Sources
50%
Confidence
Long-term
Relevance
10/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMinor version upgrades are carried out through post-training techniques such as supervised fine-tuning (SFT), reinforcement learning from human feedback (RLHF), or direct preference optimization (DPO), at a lower cost than full pretraining.77% similarUnverifiedCompared to full pre-training, SFT requires far less data and compute, but still requires a GPU cluster, a distributed training framework, and checkpoint management and experiment tracking capabilities.76% 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73% similarVerifiedDuring 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.72% similarVerified从预训练检查点到可商用产品需经历监督式指令微调（SFT）和基于人类反馈的强化学习（RLHF）两个后训练阶段72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/969751API
curl https://kongchang.com/api/v1/knowledge/claims/969751MCP
get_claim(id=969751)