Production Retrieval-Augmented Generation (RAG) pays retrieval latency and serving cost on every query – even when the model already knows the answer. We present CONFRAG, a confidence-guided strategy that invokes RAG only when a large language model (LLM) is unlikely to be correct, cutting external retrievals by 5-39% and tail latency by up to 800ms (P90) in a real RAG system while matching the answer quality of always-on RAG. CONFRAG is enabled by CONFQA, a lightweight fine-tuning strategy (one epoch over 3K knowledge-graph facts) that teaches an LLM to answer only when confident and otherwise say "I am unsure", cutting hallucination rates from 20-40% to 5% or below across multiple factuality benchmarks. Two choices make it work in practice: a dampening prompt that discourages overconfident hallucinations, and training on atomic facts that generalizes confidence calibration to unseen domains and question types. At serving time, LLM-QA and RAG run in parallel with early-stopping, so confident answers return immediately.

Large language models can generate factually inaccurate content, a problem known as hallucination. Recent works have built upon retrieved-augmented generation to improve factuality through iterative prompting but these methods are limited by the traditional RAG design. To address these challenges, we introduce Ewe (Explicit Working Memory), a novel approach that enhances factuality in long-form text generation by integrating a working memory that receives real-time feedback from external resources. The memory is refreshed based on online fact-checking and retrieval feedback, allowing Ewe to rectify false claims during the generation process and ensure more accurate and reliable outputs. Our experiments demonstrate that Ewe outperforms strong baselines on four fact-seeking long-form generation datasets, increasing the factuality metric, VeriScore, by 2 to 6 points absolute without sacrificing the helpfulness of the responses. Further analysis reveals that the design of rules for memory updates, configurations of memory units, and the quality of the retrieval datastore are crucial factors for influencing model performance.

Evaluation of biases in language models is often limited to synthetically generated datasets. This dependence traces back to the need of prompt-style dataset to trigger specific behaviors of language models. In this paper, we address this gap by creating a prompt dataset with respect to occupations collected from real-world natural sentences present in Wikipedia.We aim to understand the differences between using template-based prompts and natural sentence prompts when studying gender-occupation biases in language models. We find bias evaluations are very sensitiveto the design choices of template prompts, and we propose using natural sentence prompts as a way of more systematically using real-world sentences to move away from design decisions that may bias the results.