On the limits of retrieval-augmented generation

WED 12 AUG 20261 min17 words

↑ the link

The clearest account I have found of where RAG actually breaks, rather than where it looks impressive.

arxiv.org
0%

Saved mostly for the failure taxonomy in section 4, which matches what I keep hitting in production.

↑ was it any good?

↑ pass it on

joe@joeee.dev:~/comments$ tail -f0 entries

reading…

2000 characters left

Comments are moderated, so yours will not show up straight away.