Search-Grounded Generation
Explores methods that use search agents to retrieve, verify, and ground references for generative models, improving factual accuracy and recency.
17 papers
Where this stands
The written synthesis of this thread is for subscribers. Subscribe.
How this thread developed
Oct 2026 · Renmin University of China, Tencent
Search agents that locate, verify and crop references improve image generation for recent events.
Introduces Savanna benchmark and search-agent method to retrieve and verify visual references for image generation on recent events.
Oct 2026 · MBZUAI, Inception42
RAG's retrieval, not reading, is the bottleneck for long-context question answering
Provides evidence that retrieval quality, not reading, limits grounded generation, and proposes a joint metric for correctness and grounding.
5 further papers
Oct 2026 · Stanford University, Duke University
Most language models rarely acknowledge information that influences their answers
Trains models to self-attribute when answers depend on external information, improving grounding.
1 further paper
Oct 2026 · Adobe Research, UC San Diego
AI image editors get better results comparing alternatives than repeatedly refining
uses alternative generation and comparison as a search strategy to improve image editing results
Oct 2026 · Northwestern University
AI health overviews cite fewer top sources and often lack evidence
Evaluates source grounding and accuracy of AI health overviews compared to traditional search results.
1 further paper
Oct 2026
Language model that indexes 200 million web pages outperforms embedding search for navigation
Shows that a language model storing web page identifiers can outperform embedding search for grounded web navigation.
3 further papers
Nov 2026 · securagen.ai
Tampered embeddings hijack offline AI retrieval despite tolerance checks
Exposes a new vulnerability in offline RAG: integrity of precomputed embeddings is not verified, enabling adversarial hijacking even when tolerance checks pass.
7 of 17 papers shown