Scholar Feed

Retrieval-augmented generation

Context-Agnostic Attack

Superseded baseline#122 of 1,179 most-superseded

Superseded — cited as a baseline and beaten by newer methods

1 papers critique it · 1 beat it on benchmarks

What papers say

Verbatim critique sentences, each from a paper that cites Context-Agnostic Attack as a baseline.

a fundamental limitation shared by these existing approaches is their reliance on the assumption that an adversary can directly manipulate a user's input query. This premise is often impractical in real-world scenarios, as it confines the attack's impact to the specific, manipulated query initiated by the attacker.
— Inference Cost Attacks for Retrieval-Augmented Large Language Models

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Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.