Long-context / context-window extension
RAG
Superseded baseline#18 of 53 most-superseded
Superseded — cited as a baseline and beaten by newer methods
2 papers critique it · 2 beat it on benchmarks
What papers say
Verbatim critique sentences, each from a paper that cites RAG as a baseline.
RAG's filtering mechanism is inherently incapable of exploiting interdependencies across the entire context, limiting its effectiveness on tasks that require complex reasoning.
“RAG ... can mitigate the challenges mentioned above but introduce issues such as information loss for inappropriate extraction”
Beaten on benchmarks
Head-to-head results where a newer method reports beating RAG. Values are copied from the source paper's tables — verify against the cited paper.
XpandA beats RAG
20.6 vs 1.4
Code.Run · [Llama3.1-7B-Inst]
Long Context Scaling: Divide and Conquer via Multi-Agent Question-driven CollaborationGoA beats RAG
48.67 vs 43.18
Average F1 · [Llama 3.1 8B, 2K context]
Graph of Agents: Principled Long Context Modeling by Emergent Multi-Agent Collaboration
What to use instead
Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.