We critically assess the efficacy of the current SOTA in hallucination detection and find that its performance on the RAGTruth dataset is largely driven by a spurious correlation with data. Controlling for this effect, state-of-the-art performs no better than supervised linear probes, while requiring extensive hyperparameter tuning across datasets. Out-of-distribution generalization is currently out of reach, with all of the analyzed methods performing close to random. We propose a set of guidelines for hallucination detection and its evaluation.

Conversational agents powered by large language models (LLMs) are rapidly becoming integral to our daily interactions, generating unprecedented amounts of conversational data. Such datasets offer a powerful lens into societal interests, trending topics, and collective concerns. Yet, existing approaches typically treat these interactions as independent and miss critical insights that could emerge from aggregating and reasoning across large-scale conversation logs. In this paper, we introduce Aggregative Question Answering, a novel task requiring models to reason explicitly over thousands of user-chatbot interactions to answer aggregative queries, such as identifying emerging concerns among specific demographics. To enable research in this direction, we construct a benchmark, WildChat-AQA, comprising 6,027 aggregative questions derived from 182,330 real-world chatbot conversations. Experiments show that existing methods either struggle to reason effectively or incur prohibitive computational costs, underscoring the need for new approaches capable of extracting collective insights from large-scale conversational data.

We study the efficacy of fine-tuning Large Language Models (LLMs) for the specific task of report (government archives, news, intelligence reports) summarization. While this topic is being very actively researched - our specific application set-up faces two challenges: (i) ground-truth summaries maybe unavailable (e.g., for government archives), and (ii) availability of limited compute power - the sensitive nature of the application requires that computation is performed on-premise and for most of our experiments we use one or two A100 GPU cards. Under this set-up we conduct experiments to answer the following questions. First, given that fine-tuning the LLMs can be resource intensive, is it feasible to fine-tune them for improved report summarization capabilities on-premise? Second, what are the metrics we could leverage to assess the quality of these summaries? We conduct experiments on two different fine-tuning approaches in parallel and our findings reveal interesting trends regarding the utility of fine-tuning LLMs. Specifically, we find that in many cases, fine-tuning helps improve summary quality and in other cases it helps by reducing the number of invalid or garbage summaries.

Retrieval Augmented Generation (RAG) systems are a widespread application of Large Language Models (LLMs) in the industry. While many tools exist empowering developers to build their own systems, measuring their performance locally, with datasets reflective of the system's use cases, is a technological challenge. Solutions to this problem range from non-specific and cheap (most public datasets) to specific and costly (generating data from local documents). In this paper, we show that using public question and answer (Q&A) datasets to assess retrieval performance can lead to non-optimal systems design, and that common tools for RAG dataset generation can lead to unbalanced data. We propose solutions to these issues based on the characterization of RAG datasets through labels and through label-targeted data generation. Finally, we show that fine-tuned small LLMs can efficiently generate Q&A datasets. We believe that these observations are invaluable to the know-your-data step of RAG systems development.

The proliferation of large language models (LLMs) has significantly enhanced text generation capabilities across various industries. However, these models' ability to generate human-like text poses substantial challenges in discerning between human and AI authorship. Despite the effectiveness of existing AI-generated text detectors, their development is hindered by the lack of comprehensive, publicly available benchmarks. Current benchmarks are limited to specific scenarios, such as question answering and text polishing, and predominantly focus on English texts, failing to capture the diverse applications and linguistic nuances of LLMs. To address these limitations, this paper constructs a comprehensive bilingual benchmark in both Chinese and English to evaluate mainstream AI-generated text detectors. We categorize LLM text generation into five distinct operations: Create, Update, Delete, Rewrite, and Translate (CUDRT), encompassing all current LLMs activities. We also establish a robust benchmark evaluation framework to support scalable and reproducible experiments. For each CUDRT category, we have developed extensive datasets to thoroughly assess detector performance. By employing the latest mainstream LLMs specific to each language, our datasets provide a thorough evaluation environment. Extensive experimental results offer critical insights for optimizing AI-generated text detectors and suggest future research directions to improve detection accuracy and generalizability across various scenarios.

Verifiable generation requires large language models (LLMs) to cite source documents supporting their outputs, thereby improve output transparency and trustworthiness. Yet, previous work mainly targets the generation of sentence-level citations, lacking specificity about which parts of a sentence are backed by the cited sources. This work studies verifiable generation with subsentence-level fine-grained citations for more precise location of generated content supported by the cited sources. We first present a dataset, SCiFi, comprising 10K Wikipedia paragraphs with subsentence-level citations. Each paragraph is paired with a set of candidate source documents for citation and a query that triggers the generation of the paragraph content. On SCiFi, we evaluate the performance of state-of-the-art LLMs and strategies for processing long documents designed for these models. Our experiment results reveals key factors that could enhance the quality of citations, including the expansion of the source documents' context accessible to the models and the implementation of specialized model tuning.

We investigate whether general-domain large language models such as GPT-4 Turbo can perform risk stratification and predict post-operative outcome measures using a description of the procedure and a patient's clinical notes derived from the electronic health record. We examine predictive performance on 8 different tasks: prediction of ASA Physical Status Classification, hospital admission, ICU admission, unplanned admission, hospital mortality, PACU Phase 1 duration, hospital duration, and ICU duration. Few-shot and chain-of-thought prompting improves predictive performance for several of the tasks. We achieve F1 scores of 0.50 for ASA Physical Status Classification, 0.81 for ICU admission, and 0.86 for hospital mortality. Performance on duration prediction tasks were universally poor across all prompt strategies. Current generation large language models can assist clinicians in perioperative risk stratification on classification tasks and produce high-quality natural language summaries and explanations.

However, their capabilities on conversational summarization remains under explored. In this work we evaluate LLMs ( 10 billion parameters) on conversational summarization and showcase their performance on various prompts. We show that the summaries generated by models depend on the instructions and the performance of LLMs vary with different instructions sometimes resulting steep drop in ROUGE scores if prompts are not selected carefully. We also evaluate the models with human evaluations and discuss the limitations of the models on conversational summarization.