DeepStress: Stress-Testing Deep Search Agents
DeepStress tests search agents against unreliable, irrelevant, and false evidence, revealing why accuracy alone misses overconfident failures and wasted computation.
ls ./papers | grep "#retrieval-augmented generation"
Trending AI research papers from Hugging Face, summarized in plain language: what each paper does, why it matters and where to read more.
7 papers · #retrieval-augmented generation
DeepStress tests search agents against unreliable, irrelevant, and false evidence, revealing why accuracy alone misses overconfident failures and wasted computation.
A new reinforcement-learning approach teaches language models to study textbooks while solving unfamiliar proofs, improving transfer beyond fixed retrieval.
StreamDuo pairs a live game commentator with a wiki research agent, improving entity recognition on unseen games while reducing hallucinated names from its training game.
DeFT-RAG tackles multi-hop RAG failures with self-contained sub-questions, structured fact tracking, lightweight decisions, and reusable answers.
Glance turns the RAG generator into its own reranker, sharing one passage prefill to cut duplicated model work while improving retrieval and downstream QA in early tests.
SearchHorizonBench introduces a benchmark to evaluate when additional search is worth its cost in AI agents, connecting prediction accuracy, gain estimation, and decision utility across 45,000+ executions.
Memory systems in AI agents can both help and hinder performance compared to naive retrieval. He et al. dissect this net value into memory benefit and memory tax, revealing hidden costs and…