Saved in:
| Main Author: | |
|---|---|
| Format: | Recurso digital |
| Language: | |
| Published: |
Zenodo
2026
|
| Online Access: | https://doi.org/10.5281/zenodo.19855440 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Table of Contents:
- Current RAG evaluation conflates citation correctness (does the cited doc support the claim?) with citation faithfulness (did the model actually rely on the doc, or post-rationalize from parametric memory?). A feasible prototype would use causal interventions like context ablation and attention attribution on existing RAG benchmarks to score faithfulness, since standard metrics like RAGAS miss this distinction.