Our data and benchmarking results are available at: http://lmexam.xlore.cn.

Assessing factuality of text generated by large language models (LLMs) is an emerging yet crucial research area, aimed at alerting users to potential errors and guiding the development of more reliable LLMs. Nonetheless, the evaluators assessing factuality necessitate suitable evaluation themselves to gauge progress and foster advancements. This direction remains under-explored, resulting in substantial impediments to the progress of factuality evaluators. To mitigate this issue, we introduce a benchmark for Factuality Evaluation of large Language Models, referred to as FELM. In this benchmark, we collect responses generated from LLMs and annotate factuality labels in a fine-grained manner. Contrary to previous studies that primarily concentrate on the factuality of world knowledge (e.g.

While Large Language Models have transformed how we interact with AI systems, they suffer from a critical flaw: they confidently generate false information that sounds entirely plausible. This hallucination problem has become a major barrier to deploying these models in real-world applications where accuracy matters. We developed a fact-verification framework that catches and corrects these errors in real-time by cross-checking LLM outputs against multiple knowledge sources. Our system combines structured databases, live web searches, and academic literature to verify factual claims as they're generated. When we detect inconsistencies, we automatically correct them while preserving the natural flow of the response. Testing across various domains showed we could reduce hallucinations by 67% without sacrificing response quality. Domain experts in healthcare, finance, and scientific research rated our corrected outputs 89% satisfactory--a significant improvement over unverified LLM responses. This work offers a practical solution for making LLMs more trustworthy in applications where getting facts wrong isn't an option.

Language models often generate factually incorrect information unsupported by their training data, a phenomenon known as extrinsic hallucination. Existing mitigation approaches often degrade performance on open-ended generation and downstream tasks, limiting their practical utility. We propose an online reinforcement learning method using a novel binary retrieval-augmented reward (RAR) to address this tradeoff. Unlike continuous reward schemes, our approach assigns a reward of one only when the model's output is entirely factually correct, and zero otherwise. We evaluate our method on Qwen3 reasoning models across diverse tasks. For open-ended generation, binary RAR achieves a 39.3% reduction in hallucination rates, substantially outperforming both supervised training and continuous-reward RL baselines. In short-form question answering, the model learns calibrated abstention, strategically outputting "I don't know" when faced with insufficient parametric knowledge. This yields 44.4% and 21.7% fewer incorrect answers on PopQA and GPQA, respectively. Crucially, these factuality gains come without performance degradation on instruction following, math, or code, whereas continuous-reward RL, despite improving factuality, induces quality regressions.

Pretrained language models (LMs) are susceptible to generate text with nonfac-tual information. In this work, we measure and improve the factual accuracy of large-scale LMs for open-ended text generation.

We present a novel framework addressing a critical vulnerability in Large Language Models (LLMs): the prevalence of factual inaccuracies within intermediate reasoning steps despite correct final answers. This phenomenon poses substantial risks in high-stakes domains including healthcare, legal analysis, and scientific research, where erroneous yet confidently presented reasoning can mislead users into dangerous decisions. Our framework integrates three core components: (1) a specialized fact-checking classifier trained on counterfactually augmented data to detect subtle factual inconsistencies within reasoning chains; (2) an enhanced Group Relative Policy Optimization (GRPO) reinforcement learning approach that balances factuality, coherence, and structural correctness through multi-dimensional rewards; and (3) a mechanistic interpretability method examining how factuality improvements manifest in model activations during reasoning processes. Extensive evaluation across multi state-of-the-art models reveals concerning patterns: even leading models like Claude-3.7 and GPT-o1 demonstrate reasoning factual accuracy of only 81.93% and 82.57% respectively. Our approach significantly enhances factual robustness (up to 49.90% improvement) while maintaining or improving performance on challenging benchmarks including Math-500, AIME-2024, and GPQA. Furthermore, our neural activation-level analysis provides actionable insights into how factual enhancements reshape reasoning trajectories within model architectures, establishing foundations for future training methodologies that explicitly target factual robustness through activation-guided optimization.