While large language models (LLMs) have shown remarkable capabilities to generate coherent text, they suffer from the issue of hallucinations -- factually inaccurate statements. Among numerous approaches to tackle hallucinations, especially promising are the self-correcting methods. They leverage the multi-turn nature of LLMs to iteratively generate verification questions inquiring additional evidence, answer them with internal or external knowledge, and use that to refine the original response with the new corrections. These methods have been explored for encyclopedic generation, but less so for domains like news summarization. In this work, we investigate two state-of-the-art self-correcting systems by applying them to correct hallucinated summaries using evidence from three search engines. We analyze the results and provide insights into systems' performance, revealing interesting practical findings on the benefits of search engine snippets and few-shot prompts, as well as high alignment of G-Eval and human evaluation.

We introduce xbench, a dynamic, profession-aligned evaluation suite designed to bridge the gap between AI agent capabilities and real-world productivity. While existing benchmarks often focus on isolated technical skills, they may not accurately reflect the economic value agents deliver in professional settings. To address this, xbench targets commercially significant domains with evaluation tasks defined by industry professionals. Our framework creates metrics that strongly correlate with productivity value, enables prediction of Technology-Market Fit (TMF), and facilitates tracking of product capabilities over time. As our initial implementations, we present two benchmarks: Recruitment and Marketing. For Recruitment, we collect 50 tasks from real-world headhunting business scenarios to evaluate agents' abilities in company mapping, information retrieval, and talent sourcing. For Marketing, we assess agents' ability to match influencers with advertiser needs, evaluating their performance across 50 advertiser requirements using a curated pool of 836 candidate influencers. We present initial evaluation results for leading contemporary agents, establishing a baseline for these professional domains. Our continuously updated evalsets and evaluations are available at https://xbench.org.

Multi-answer question answering (QA), where questions can have many valid answers, presents a significant challenge for existing retrieval-augmented generation-based QA systems, as these systems struggle to retrieve and then synthesize a large number of evidence passages. To tackle these challenges, we propose a new multi-answer QA framework -- Retrieval-augmented Independent Reading with Inter-passage Verification (RI$^2$VER). Our framework retrieves a large set of passages and processes each passage individually to generate an initial high-recall but noisy answer set. Then we propose a new inter-passage verification pipeline that validates every candidate answer through (1) Verification Question Generation, (2) Gathering Additional Evidence, and (3) Verification with inter-passage synthesis. Evaluations on the QAMPARI and RoMQA datasets demonstrate that our framework significantly outperforms existing baselines across various model sizes, achieving an average F1 score improvement of 11.17%. Further analysis validates that our inter-passage verification pipeline enables our framework to be particularly beneficial for questions requiring multi-evidence synthesis.

Large language models (LLMs) are increasingly integrated into real-world applications through retrieval-augmented generation (RAG) mechanisms to supplement their responses with up-to-date and domain-specific knowledge. However, the valuable and often proprietary nature of the knowledge bases used in RAG introduces the risk of unauthorized usage by adversaries. Existing methods that can be generalized as watermarking techniques to protect these knowledge bases typically involve poisoning attacks. However, these methods require to alter the results of verification samples (\eg, generating incorrect outputs), inevitably making them susceptible to anomaly detection and even introduce new security risks. To address these challenges, we propose \name{} for `harmless' copyright protection of knowledge bases. Instead of manipulating LLM's final output, \name{} implants distinct verification behaviors in the space of chain-of-thought (CoT) reasoning, maintaining the correctness of the final answer. Our method has three main stages: (1) \textbf{Generating CoTs}: For each verification question, we generate two CoTs, including a target CoT for building watermark behaviors; (2) \textbf{Optimizing Watermark Phrases and Target CoTs}: We optimize them to minimize retrieval errors under the black-box setting of suspicious LLM, ensuring that the watermarked verification queries activate the target CoTs without being activated in non-watermarked ones; (3) \textbf{Ownership Verification}: We exploit a pairwise Wilcoxon test to statistically verify whether a suspicious LLM is augmented with the protected knowledge base by comparing its responses to watermarked and benign verification queries. Our experiments on diverse benchmarks demonstrate that \name{} effectively protects knowledge bases against unauthorized usage while preserving the integrity and performance of the RAG.

Large Language Models (LLMs) are powerful computational models trained on extensive corpora of human-readable text, enabling them to perform general-purpose language understanding and generation. LLMs have garnered significant attention in both industry and academia due to their exceptional performance across various natural language processing (NLP) tasks. Despite these successes, LLMs often produce inaccuracies, commonly referred to as hallucinations. Prompt engineering, the process of designing and formulating instructions for LLMs to perform specific tasks, has emerged as a key approach to mitigating hallucinations. This paper provides a comprehensive empirical evaluation of different prompting strategies and frameworks aimed at reducing hallucinations in LLMs. Various prompting techniques are applied to a broad set of benchmark datasets to assess the accuracy and hallucination rate of each method. Additionally, the paper investigates the influence of tool-calling agents (LLMs augmented with external tools to enhance their capabilities beyond language generation) on hallucination rates in the same benchmarks. The findings demonstrate that the optimal prompting technique depends on the type of problem, and that simpler techniques often outperform more complex methods in reducing hallucinations. Furthermore, it is shown that LLM agents can exhibit significantly higher hallucination rates due to the added complexity of external tool usage.

Large Language Models (LLMs) show promise for natural language generation in healthcare, but risk hallucinating factually incorrect information. Deploying LLMs for medical question answering necessitates reliable uncertainty estimation (UE) methods to detect hallucinations. In this work, we benchmark popular UE methods with different model sizes on medical question-answering datasets. Our results show that current approaches generally perform poorly in this domain, highlighting the challenge of UE for medical applications. We also observe that larger models tend to yield better results, suggesting a correlation between model size and the reliability of UE. To address these challenges, we propose Two-phase Verification, a probability-free Uncertainty Estimation approach. First, an LLM generates a step-by-step explanation alongside its initial answer, followed by formulating verification questions to check the factual claims in the explanation. The model then answers these questions twice: first independently, and then referencing the explanation. Inconsistencies between the two sets of answers measure the uncertainty in the original response. We evaluate our approach on three biomedical question-answering datasets using Llama 2 Chat models and compare it against the benchmarked baseline methods. The results show that our Two-phase Verification method achieves the best overall accuracy and stability across various datasets and model sizes, and its performance scales as the model size increases.

Intrinsic self-correct was a method that instructed large language models (LLMs) to verify and correct their responses without external feedback. Unfortunately, the study concluded that the LLMs could not self-correct reasoning yet. We find that a simple yet effective verification method can unleash inherent capabilities of the LLMs. That is to mask a key condition in the question, add the current response to construct a verification question, and predict the condition to verify the response. The condition can be an entity in an open-domain question or a numeric value in a math question, which requires minimal effort (via prompting) to identify. We propose an iterative verify-then-correct framework to progressively identify and correct (probably) false responses, named ProCo. We conduct experiments on three reasoning tasks. On average, ProCo, with GPT-3.5-Turbo as the backend LLM, yields $+6.8$ exact match on four open-domain question answering datasets, $+14.1$ accuracy on three arithmetic reasoning datasets, and $+9.6$ accuracy on a commonsense reasoning dataset, compared to Self-Correct.

Teaching language models to use tools is an important milestone towards building general assistants, but remains an open problem. While there has been significant progress on learning to use specific tools via fine-tuning, language models still struggle with learning how to robustly use new tools from only a few demonstrations. In this work we introduce a self-verification method which distinguishes between close candidates by self-asking contrastive questions during (1) tool selection; and (2) parameter generation. We construct synthetic, high-quality, self-generated data for this goal using Llama-2 70B, which we intend to release publicly. Extensive experiments on 4 tasks from the ToolBench benchmark, consisting of 17 unseen tools, demonstrate an average improvement of 22% over few-shot baselines, even in scenarios where the distinctions between candidate tools are finely nuanced.

Generation of plausible yet incorrect factual information, termed hallucination, is an unsolved issue in large language models. We study the ability of language models to deliberate on the responses they give in order to correct their mistakes. We develop the Chain-of-Verification (CoVe) method whereby the model first (i) drafts an initial response; then (ii) plans verification questions to fact-check its draft; (iii) answers those questions independently so the answers are not biased by other responses; and (iv) generates its final verified response. In experiments, we show CoVe decreases hallucinations across a variety of tasks, from list-based questions from Wikidata, closed book MultiSpanQA and longform text generation.

Much recent research has uncovered and discussed serious concerns of bias in facial analysis technologies, finding performance disparities between groups of people based on perceived gender, skin type, lighting condition, etc. These audits are immensely important and successful at measuring algorithmic bias but have two major challenges: the audits (1) use facial recognition datasets which lack quality metadata, like LFW and CelebA, and (2) do not compare their observed algorithmic bias to the biases of their human alternatives. In this paper, we release improvements to the LFW and CelebA datasets which will enable future researchers to obtain measurements of algorithmic bias that are not tainted by major flaws in the dataset (e.g. identical images appearing in both the gallery and test set). We also use these new data to develop a series of challenging facial identification and verification questions that we administered to various algorithms and a large, balanced sample of human reviewers. We find that both computer models and human survey participants perform significantly better at the verification task, generally obtain lower accuracy rates on dark-skinned or female subjects for both tasks, and obtain higher accuracy rates when their demographics match that of the question. Computer models are observed to achieve a higher level of accuracy than the survey participants on both tasks and exhibit bias to similar degrees as the human survey participants.