Unverified50% confidenceOpinionExact time
Multi-hop question answering requires chained reasoning across multiple documents or passages; traditional RAG fixed pipelines often lose context or introduce noise after the second hop, whereas RPL allows the model to explicitly write loops and conditional branches to advance hop by hop.
1
Sources
50%
Confidence
Long-term
Relevance
10/20/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedFine-tuning is suitable for teaching a model a specific output style or domain tone, while RAG achieves immediate-update-ready results through an external knowledge base, and the two are often used in combination in engineering.70% similarUnverifiedPrompt engineering, RAG, and fine-tuning apply to different scenarios: prompt engineering acts on a single inference, is low-cost, but its rules are easily ignored and it cannot cross sessions; RAG's external knowledge base is suitable for frequently updated factual knowledge but does not change the model's reasoning habits; fine-tuning modifies the weights so that the model acquires consistent behavioral patterns.69% similarUnverified提示词与模型版本存在深度绑定关系，不同版本模型在RLHF训练数据、指令微调策略上的差异会导致相同提示词产生不同输出，这种现象被称为「提示词漂移」67% similarVerifiedEach RL training step requires running inference (generating rollouts) and backpropagation (updating weights) simultaneously, making the compute requirements far higher than ordinary fine-tuning.67% similarUnverified三种RAG路线并非互斥，实际工程中可采用混合架构：用向量检索做初步召回，用图谱结构做关系推理，用编译层做知识管理和溯源66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1074711API
curl https://kongchang.com/api/v1/knowledge/claims/1074711MCP
get_claim(id=1074711)