We present PeerQA, a real-world, scientific, document-level Question Answering (QA) dataset. PeerQA questions have been sourced from peer reviews, which contain questions that reviewers raised while thoroughly examining the scientific article. Answers have been annotated by the original authors of each paper. The dataset contains 579 QA pairs from 208 academic articles, with a majority from ML and NLP, as well as a subset of other scientific communities like Geoscience and Public Health. PeerQA supports three critical tasks for developing practical QA systems: Evidence retrieval, unanswerable question classification, and answer generation. We provide a detailed analysis of the collected dataset and conduct experiments establishing baseline systems for all three tasks. Our experiments and analyses reveal the need for decontextualization in document-level retrieval, where we find that even simple decontextualization approaches consistently improve retrieval performance across architectures. On answer generation, PeerQA serves as a challenging benchmark for long-context modeling, as the papers have an average size of 12k tokens. Our code and data is available at https://github.com/UKPLab/peerqa.
Modern artificial intelligence (AI) systems are powered by foundation models. This paper presents a new set of foundation models, called Llama 3. It is a herd of language models that natively support multilinguality, coding, reasoning, and tool usage. Our largest model is a dense Transformer with 405B parameters and a context window of up to 128K tokens. This paper presents an extensive empirical evaluation of Llama 3. We find that Llama 3 delivers comparable quality to leading language models such as GPT-4 on a plethora of tasks. We publicly release Llama 3, including pre-trained and post-trained versions of the 405B parameter language model and our Llama Guard 3 model for input and output safety. The paper also presents the results of experiments in which we integrate image, video, and speech capabilities into Llama 3 via a compositional approach. We observe this approach performs competitively with the state-of-the-art on image, video, and speech recognition tasks. The resulting models are not yet being broadly released as they are still under development.
Generative AI models face the challenge of hallucinations that can undermine users' trust in such systems. We approach the problem of conversational information seeking as a two-step process, where relevant passages in a corpus are identified first and then summarized into a final system response. This way we can automatically assess if the answer to the user's question is present in the corpus. Specifically, our proposed method employs a sentence-level classifier to detect if the answer is present, then aggregates these predictions on the passage level, and eventually across the top-ranked passages to arrive at a final answerability estimate. For training and evaluation, we develop a dataset based on the TREC CAsT benchmark that includes answerability labels on the sentence, passage, and ranking levels. We demonstrate that our proposed method represents a strong baseline and outperforms a state-of-the-art LLM on the answerability prediction task.
We introduce Mistral 7B v0.1, a 7-billion-parameter language model engineered for superior performance and efficiency. Mistral 7B outperforms Llama 2 13B across all evaluated benchmarks, and Llama 1 34B in reasoning, mathematics, and code generation. Our model leverages grouped-query attention (GQA) for faster inference, coupled with sliding window attention (SWA) to effectively handle sequences of arbitrary length with a reduced inference cost. We also provide a model fine-tuned to follow instructions, Mistral 7B -- Instruct, that surpasses the Llama 2 13B -- Chat model both on human and automated benchmarks. Our models are released under the Apache 2.0 license.
We report the development of GPT-4, a large-scale, multimodal model which can accept image and text inputs and produce text outputs. While less capable than humans in many real-world scenarios, GPT-4 exhibits human-level performance on various professional and academic benchmarks, including passing a simulated bar exam with a score around the top 10% of test takers. GPT-4 is a Transformer-based model pre-trained to predict the next token in a document. The post-training alignment process results in improved performance on measures of factuality and adherence to desired behavior. A core component of this project was developing infrastructure and optimization methods that behave predictably across a wide range of scales. This allowed us to accurately predict some aspects of GPT-4's performance based on models trained with no more than 1/1,000th the compute of GPT-4.
There has been considerable progress on academic benchmarks for the Reading Comprehension (RC) task with State-of-the-Art models closing the gap with human performance on extractive question answering. Datasets such as SQuAD 2.0 & NQ have also introduced an auxiliary task requiring models to predict when a question has no answer in the text. However, in production settings, it is also necessary to provide confidence estimates for the performance of the underlying RC model at both answer extraction and "answerability" detection. We propose a novel post-prediction confidence estimation model, which we call Mr.C (short for Mr. Confident), that can be trained to improve a system's ability to refrain from making incorrect predictions with improvements of up to 4 points as measured by Area Under the Curve (AUC) scores. Mr.C can benefit from a novel white-box feature that leverages the underlying RC model's gradients. Performance prediction is particularly important in cases of domain shift (as measured by training RC models on SQUAD 2.0 and evaluating on NQ), where Mr.C not only improves AUC, but also traditional answerability prediction (as measured by a 5 point improvement in F1).
Recent pretrained language models "solved" many reading comprehension benchmarks, where questions are written with access to the evidence document. However, datasets containing information-seeking queries where evidence documents are provided after the queries are written independently remain challenging. We analyze why answering information-seeking queries is more challenging and where their prevalent unanswerabilities arise, on Natural Questions and TyDi QA. Our controlled experiments suggest two headrooms -- paragraph selection and answerability prediction, i.e. whether the paired evidence document contains the answer to the query or not. When provided with a gold paragraph and knowing when to abstain from answering, existing models easily outperform a human annotator. However, predicting answerability itself remains challenging. We manually annotate 800 unanswerable examples across six languages on what makes them challenging to answer. With this new data, we conduct per-category answerability prediction, revealing issues in the current dataset collection as well as task formulation. Together, our study points to avenues for future research in information-seeking question answering, both for dataset creation and model development.
Recent work has demonstrated substantial gains on many NLP tasks and benchmarks by pre-training on a large corpus of text followed by fine-tuning on a specific task. While typically task-agnostic in architecture, this method still requires task-specific fine-tuning datasets of thousands or tens of thousands of examples. By contrast, humans can generally perform a new language task from only a few examples or from simple instructions - something which current NLP systems still largely struggle to do. Here we show that scaling up language models greatly improves task-agnostic, few-shot performance, sometimes even reaching competitiveness with prior state-of-the-art fine-tuning approaches. Specifically, we train GPT-3, an autoregressive language model with 175 billion parameters, 10x more than any previous non-sparse language model, and test its performance in the few-shot setting. For all tasks, GPT-3 is applied without any gradient updates or fine-tuning, with tasks and few-shot demonstrations specified purely via text interaction with the model. GPT-3 achieves strong performance on many NLP datasets, including translation, question-answering, and cloze tasks, as well as several tasks that require on-the-fly reasoning or domain adaptation, such as unscrambling words, using a novel word in a sentence, or performing 3-digit arithmetic. At the same time, we also identify some datasets where GPT-3's few-shot learning still struggles, as well as some datasets where GPT-3 faces methodological issues related to training on large web corpora. Finally, we find that GPT-3 can generate samples of news articles which human evaluators have difficulty distinguishing from articles written by humans. We discuss broader societal impacts of this finding and of GPT-3 in general.