Philippe Laban

Erik Nijkamp, Tian Xie, Hiroaki Hayashi et al. · cmu, microsoft-research

Large Language Models (LLMs) have become ubiquitous across various domains, transforming the way we interact with information and conduct research. However, most high-performing LLMs remain confined behind proprietary walls, hindering scientific progress. Most open-source LLMs, on the other hand, are limited in their ability to support longer sequence lengths, which is a key requirement for many tasks that require inference over an input context. To address this, we have trained XGen, a series of 7B parameter models on up to 8K sequence length for up to 1.5T tokens. We have also finetuned the XGen models on public-domain instructional data, creating their instruction-tuned counterparts (XGen-Inst). We open-source our models for both research advancements and commercial applications. Our evaluation on standard benchmarks shows that XGen models achieve comparable or better results when compared with state-of-the-art open-source LLMs. Our targeted evaluation on long sequence modeling tasks shows the benefits of our 8K-sequence models over 2K-sequence open-source LLMs.

Philippe Laban, Lidiya Murakhovs'ka, Caiming Xiong et al. · microsoft-research, salesforce

The interactive nature of Large Language Models (LLMs) theoretically allows models to refine and improve their answers, yet systematic analysis of the multi-turn behavior of LLMs remains limited. In this paper, we propose the FlipFlop experiment: in the first round of the conversation, an LLM completes a classification task. In a second round, the LLM is challenged with a follow-up phrase like "Are you sure?", offering an opportunity for the model to reflect on its initial answer, and decide whether to confirm or flip its answer. A systematic study of ten LLMs on seven classification tasks reveals that models flip their answers on average 46% of the time and that all models see a deterioration of accuracy between their first and final prediction, with an average drop of 17% (the FlipFlop effect). We conduct finetuning experiments on an open-source LLM and find that finetuning on synthetically created data can mitigate - reducing performance deterioration by 60% - but not resolve sycophantic behavior entirely. The FlipFlop experiment illustrates the universality of sycophantic behavior in LLMs and provides a robust framework to analyze model behavior and evaluate future models.

Liyan Tang, Tanya Goyal, Alexander R. Fabbri et al. · microsoft-research, salesforce

The propensity of abstractive summarization models to make factual errors has been studied extensively, including design of metrics to detect factual errors and annotation of errors in current systems' outputs. However, the ever-evolving nature of summarization systems, metrics, and annotated benchmarks makes factuality evaluation a moving target, and drawing clear comparisons among metrics has become increasingly difficult. In this work, we aggregate factuality error annotations from nine existing datasets and stratify them according to the underlying summarization model. We compare performance of state-of-the-art factuality metrics, including recent ChatGPT-based metrics, on this stratified benchmark and show that their performance varies significantly across different types of summarization models. Critically, our analysis shows that much of the recent improvement in the factuality detection space has been on summaries from older (pre-Transformer) models instead of more relevant recent summarization models. We further perform a finer-grained analysis per error-type and find similar performance variance across error types for different factuality metrics. Our results show that no one metric is superior in all settings or for all error types, and we provide recommendations for best practices given these insights.

Philippe Laban, Jesse Vig, Marti A. Hearst et al. · microsoft-research, salesforce

Conversational interfaces powered by Large Language Models (LLMs) have recently become a popular way to obtain feedback during document editing. However, standard chat-based conversational interfaces do not support transparency and verifiability of the editing changes that they suggest. To give the author more agency when editing with an LLM, we present InkSync, an editing interface that suggests executable edits directly within the document being edited. Because LLMs are known to introduce factual errors, Inksync also supports a 3-stage approach to mitigate this risk: Warn authors when a suggested edit introduces new information, help authors Verify the new information's accuracy through external search, and allow an auditor to perform an a-posteriori verification by Auditing the document via a trace of all auto-generated content. Two usability studies confirm the effectiveness of InkSync's components when compared to standard LLM-based chat interfaces, leading to more accurate, more efficient editing, and improved user experience.

Tuhin Chakrabarty, Philippe Laban, Divyansh Agarwal et al. · microsoft-research, salesforce

Researchers have argued that large language models (LLMs) exhibit high-quality writing capabilities from blogs to stories. However, evaluating objectively the creativity of a piece of writing is challenging. Inspired by the Torrance Test of Creative Thinking (TTCT), which measures creativity as a process, we use the Consensual Assessment Technique [3] and propose the Torrance Test of Creative Writing (TTCW) to evaluate creativity as a product. TTCW consists of 14 binary tests organized into the original dimensions of Fluency, Flexibility, Originality, and Elaboration. We recruit 10 creative writers and implement a human assessment of 48 stories written either by professional authors or LLMs using TTCW. Our analysis shows that LLM-generated stories pass 3-10X less TTCW tests than stories written by professionals. In addition, we explore the use of LLMs as assessors to automate the TTCW evaluation, revealing that none of the LLMs positively correlate with the expert assessments.

Fan Yin, Jesse Vig, Philippe Laban et al. · microsoft-research, salesforce

Large language models (LLMs) have shown impressive performance in following natural language instructions to solve unseen tasks. However, it remains unclear whether models truly understand task definitions and whether the human-written definitions are optimal. In this paper, we systematically study the role of task definitions in instruction learning. We first conduct an ablation analysis informed by human annotations to understand which parts of a task definition are most important, and find that model performance only drops substantially when removing contents describing the task output, in particular label information. Next, we propose an automatic algorithm to compress task definitions to a minimal supporting set of tokens, and find that 60\% of tokens can be removed while maintaining or even improving model performance. Based on these results, we propose two strategies to help models better leverage task instructions: (1) providing only key information for tasks in a common structured format, and (2) adding a meta-tuning stage to help the model better understand the definitions. With these two strategies, we achieve a 4.2 Rouge-L improvement over 119 unseen test tasks.

Kung-Hsiang Huang, Philippe Laban, Alexander R. Fabbri et al. · microsoft-research, salesforce

Previous research in multi-document news summarization has typically concentrated on collating information that all sources agree upon. However, the summarization of diverse information dispersed across multiple articles about an event remains underexplored. In this paper, we propose a new task of summarizing diverse information encountered in multiple news articles encompassing the same event. To facilitate this task, we outlined a data collection schema for identifying diverse information and curated a dataset named DiverseSumm. The dataset includes 245 news stories, with each story comprising 10 news articles and paired with a human-validated reference. Next, to enable consistent automatic evaluation, we conducted a comprehensive analysis to pinpoint the position and verbosity biases when utilizing Large Language Model (LLM)-based metrics for evaluating the coverage and faithfulness of summaries. Through correlation analyses, we outline the best practices for effectively using automatic LLM-based metrics on the DiverseSumm dataset. Finally, we study how LLMs summarize multiple news articles by analyzing which type of diverse information LLMs are capable of identifying. Our analyses suggest that despite the extraordinary capabilities of LLMs in single-document summarization, the proposed task remains a complex challenge for them mainly due to their limited coverage, with GPT-4 only able to cover under 40% of the diverse information on average.

Xiang 'Anthony' Chen, Jeff Burke, Ruofei Du et al. · microsoft-research, salesforce

Through iterative, cross-disciplinary discussions, we define and propose next-steps for Human-centered Generative AI (HGAI). We contribute a comprehensive research agenda that lays out future directions of Generative AI spanning three levels: aligning with human values; assimilating human intents; and augmenting human abilities. By identifying these next-steps, we intend to draw interdisciplinary research teams to pursue a coherent set of emergent ideas in HGAI, focusing on their interested topics while maintaining a coherent big picture of the future work landscape.

Philippe Laban, Alexander R. Fabbri, Caiming Xiong et al. · microsoft-research

LLMs and RAG systems are now capable of handling millions of input tokens or more. However, evaluating the output quality of such systems on long-context tasks remains challenging, as tasks like Needle-in-a-Haystack lack complexity. In this work, we argue that summarization can play a central role in such evaluation. We design a procedure to synthesize Haystacks of documents, ensuring that specific \textit{insights} repeat across documents. The "Summary of a Haystack" (SummHay) task then requires a system to process the Haystack and generate, given a query, a summary that identifies the relevant insights and precisely cites the source documents. Since we have precise knowledge of what insights should appear in a haystack summary and what documents should be cited, we implement a highly reproducible automatic evaluation that can score summaries on two aspects - Coverage and Citation. We generate Haystacks in two domains (conversation, news), and perform a large-scale evaluation of 10 LLMs and corresponding 50 RAG systems. Our findings indicate that SummHay is an open challenge for current systems, as even systems provided with an Oracle signal of document relevance lag our estimate of human performance (56\%) by 10+ points on a Joint Score. Without a retriever, long-context LLMs like GPT-4o and Claude 3 Opus score below 20% on SummHay. We show SummHay can also be used to study enterprise RAG systems and position bias in long-context models. We hope future systems can equal and surpass human performance on SummHay.

Philippe Laban, Chien-Sheng Wu, Lidiya Murakhovs'ka et al. · microsoft-research, salesforce

Question generation (QGen) models are often evaluated with standardized NLG metrics that are based on n-gram overlap. In this paper, we measure whether these metric improvements translate to gains in a practical setting, focusing on the use case of helping teachers automate the generation of reading comprehension quizzes. In our study, teachers building a quiz receive question suggestions, which they can either accept or refuse with a reason. Even though we find that recent progress in QGen leads to a significant increase in question acceptance rates, there is still large room for improvement, with the best model having only 68.4% of its questions accepted by the ten teachers who participated in our study. We then leverage the annotations we collected to analyze standard NLG metrics and find that model performance has reached projected upper-bounds, suggesting new automatic metrics are needed to guide QGen research forward.

Philippe Laban, Chien-Sheng Wu, Lidiya Murakhovs'ka et al. · microsoft-research, salesforce

There are many potential benefits to news readers accessing diverse sources. Modern news aggregators do the hard work of organizing the news, offering readers a plethora of source options, but choosing which source to read remains challenging. We propose a new framework to assist readers in identifying source differences and gaining an understanding of news coverage diversity. The framework is based on the generation of Discord Questions: questions with a diverse answer pool, explicitly illustrating source differences. To assemble a prototype of the framework, we focus on two components: (1) discord question generation, the task of generating questions answered differently by sources, for which we propose an automatic scoring method, and create a model that improves performance from current question generation (QG) methods by 5%, (2) answer consolidation, the task of grouping answers to a question that are semantically similar, for which we collect data and repurpose a method that achieves 81% balanced accuracy on our realistic test set. We illustrate the framework's feasibility through a prototype interface. Even though model performance at discord QG still lags human performance by more than 15%, generated questions are judged to be more interesting than factoid questions and can reveal differences in the level of detail, sentiment, and reasoning of sources in news coverage.

Tuhin Chakrabarty, Philippe Laban, Chien-Sheng Wu · microsoft-research

LLM-based applications are helping people write, and LLM-generated text is making its way into social media, journalism, and our classrooms. However, the differences between LLM-generated and human written text remain unclear. To explore this, we hired professional writers to edit paragraphs in several creative domains. We first found these writers agree on undesirable idiosyncrasies in LLM generated text, formalizing it into a seven-category taxonomy (e.g. clichés, unnecessary exposition). Second, we curated the LAMP corpus: 1,057 LLM-generated paragraphs edited by professional writers according to our taxonomy. Analysis of LAMP reveals that none of the LLMs used in our study (GPT4o, Claude-3.5-Sonnet, Llama-3.1-70b) outperform each other in terms of writing quality, revealing common limitations across model families. Third, building on existing work in automatic editing we evaluated methods to improve LLM-generated text. A large-scale preference annotation confirms that although experts largely prefer text edited by other experts, automatic editing methods show promise in improving alignment between LLM-generated and human-written text.

Philippe Laban, Chien-Sheng Wu, Lidiya Murakhovs'ka et al. · microsoft-research, salesforce

Modern news aggregators do the hard work of organizing a large news stream, creating collections for a given news story with tens of source options. This paper shows that navigating large source collections for a news story can be challenging without further guidance. In this work, we design three interfaces -- the Annotated Article, the Recomposed Article, and the Question Grid -- aimed at accompanying news readers in discovering coverage diversity while they read. A first usability study with 10 journalism experts confirms the designed interfaces all reveal coverage diversity and determine each interface's potential use cases and audiences. In a second usability study, we developed and implemented a reading exercise with 95 novice news readers to measure exposure to coverage diversity. Results show that Annotated Article users are able to answer questions 34% more completely than with two existing interfaces while finding the interface equally easy to use.

Lidiya Murakhovs'ka, Philippe Laban, Tian Xie et al. · microsoft-research, salesforce

Making big purchases requires consumers to research or consult a salesperson to gain domain expertise. However, existing conversational recommender systems (CRS) often overlook users' lack of background knowledge, focusing solely on gathering preferences. In this work, we define a new problem space for conversational agents that aim to provide both product recommendations and educational value through mixed-type mixed-initiative dialog. We introduce SalesOps, a framework that facilitates the simulation and evaluation of such systems by leveraging recent advancements in large language models (LLMs). We build SalesBot and ShopperBot, a pair of LLM-powered agents that can simulate either side of the framework. A comprehensive human study compares SalesBot against professional salespeople, revealing that although SalesBot approaches professional performance in terms of fluency and informativeness, it lags behind in recommendation quality. We emphasize the distinct limitations both face in providing truthful information, highlighting the challenges of ensuring faithfulness in the CRS context. We release our code and make all data available.

Philippe Laban, Chien-Sheng Wu, Wenhao Liu et al. · microsoft-research, salesforce

Precisely assessing the progress in natural language generation (NLG) tasks is challenging, and human evaluation to establish a preference in a model's output over another is often necessary. However, human evaluation is usually costly, difficult to reproduce, and non-reusable. In this paper, we propose a new and simple automatic evaluation method for NLG called Near-Negative Distinction (NND) that repurposes prior human annotations into NND tests. In an NND test, an NLG model must place a higher likelihood on a high-quality output candidate than on a near-negative candidate with a known error. Model performance is established by the number of NND tests a model passes, as well as the distribution over task-specific errors the model fails on. Through experiments on three NLG tasks (question generation, question answering, and summarization), we show that NND achieves a higher correlation with human judgments than standard NLG evaluation metrics. We then illustrate NND evaluation in four practical scenarios, for example performing fine-grain model analysis, or studying model training dynamics. Our findings suggest that NND can give a second life to human annotations and provide low-cost NLG evaluation.

Yao Dou, Philippe Laban, Claire Gardent et al. · microsoft-research

In this tutorial, we focus on text-to-text generation, a class of natural language generation (NLG) tasks, that takes a piece of text as input and then generates a revision that is improved according to some specific criteria (e.g., readability or linguistic styles), while largely retaining the original meaning and the length of the text. This includes many useful applications, such as text simplification, paraphrase generation, style transfer, etc. In contrast to text summarization and open-ended text completion (e.g., story), the text-to-text generation tasks we discuss in this tutorial are more constrained in terms of semantic consistency and targeted language styles. This level of control makes these tasks ideal testbeds for studying the ability of models to generate text that is both semantically adequate and stylistically appropriate. Moreover, these tasks are interesting from a technical standpoint, as they require complex combinations of lexical and syntactical transformations, stylistic control, and adherence to factual knowledge, -- all at once. With a special focus on text simplification and revision, this tutorial aims to provide an overview of the state-of-the-art natural language generation research from four major aspects -- Data, Models, Human-AI Collaboration, and Evaluation -- and to discuss and showcase a few significant and recent advances: (1) the use of non-retrogressive approaches; (2) the shift from fine-tuning to prompting with large language models; (3) the development of new learnable metric and fine-grained human evaluation framework; (4) a growing body of studies and datasets on non-English languages; (5) the rise of HCI+NLP+Accessibility interdisciplinary research to create real-world writing assistant systems.

Divyansh Agarwal, Alexander R. Fabbri, Ben Risher et al. · microsoft-research

Prompt leakage poses a compelling security and privacy threat in LLM applications. Leakage of system prompts may compromise intellectual property, and act as adversarial reconnaissance for an attacker. A systematic evaluation of prompt leakage threats and mitigation strategies is lacking, especially for multi-turn LLM interactions. In this paper, we systematically investigate LLM vulnerabilities against prompt leakage for 10 closed- and open-source LLMs, across four domains. We design a unique threat model which leverages the LLM sycophancy effect and elevates the average attack success rate (ASR) from 17.7% to 86.2% in a multi-turn setting. Our standardized setup further allows dissecting leakage of specific prompt contents such as task instructions and knowledge documents. We measure the mitigation effect of 7 black-box defense strategies, along with finetuning an open-source model to defend against leakage attempts. We present different combination of defenses against our threat model, including a cost analysis. Our study highlights key takeaways for building secure LLM applications and provides directions for research in multi-turn LLM interactions

Philippe Laban, Tobias Schnabel, Jennifer Neville

Large Language Models (LLMs) are poised to disrupt knowledge work, with the emergence of delegated work as a new interaction paradigm (e.g., vibe coding). Delegation requires trust - the expectation that the LLM will faithfully execute the task without introducing errors into documents. We introduce DELEGATE-52 to study the readiness of AI systems in delegated workflows. DELEGATE-52 simulates long delegated workflows that require in-depth document editing across 52 professional domains, such as coding, crystallography, and music notation. Our large-scale experiment with 19 LLMs reveals that current models degrade documents during delegation: even frontier models (Gemini 3.1 Pro, Claude 4.6 Opus, GPT 5.4) corrupt an average of 25% of document content by the end of long workflows, with other models failing more severely. Additional experiments reveal that agentic tool use does not improve performance on DELEGATE-52, and that degradation severity is exacerbated by document size, length of interaction, or presence of distractor files. Our analysis shows that current LLMs are unreliable delegates: they introduce sparse but severe errors that silently corrupt documents, compounding over long interaction.

Liyan Tang, Philippe Laban, Greg Durrett · microsoft-research

Recognizing if LLM output can be grounded in evidence is central to many tasks in NLP: retrieval-augmented generation, summarization, document-grounded dialogue, and more. Current approaches to this kind of fact-checking are based on verifying each piece of a model generation against potential evidence using an LLM. However, this process can be very computationally expensive, requiring many calls to a model to check a single response. In this work, we show how to build small fact-checking models that have GPT-4-level performance but for 400x lower cost. We do this by constructing synthetic training data with GPT-4, which involves creating realistic yet challenging instances of factual errors via a structured generation procedure. Training on this data teaches models to check each fact in the claim and recognize synthesis of information across sentences. For evaluation, we unify datasets from recent work on fact-checking and grounding LLM generations into a new benchmark, LLM-AggreFact. Our best system MiniCheck-FT5 (770M parameters) outperforms all systems of comparable size and reaches GPT-4 accuracy. We release LLM-AggreFact, code for data synthesis, and models.

Philippe Laban, Tobias Schnabel, Paul N. Bennett et al.

In the summarization domain, a key requirement for summaries is to be factually consistent with the input document. Previous work has found that natural language inference (NLI) models do not perform competitively when applied to inconsistency detection. In this work, we revisit the use of NLI for inconsistency detection, finding that past work suffered from a mismatch in input granularity between NLI datasets (sentence-level), and inconsistency detection (document level). We provide a highly effective and light-weight method called SummaCConv that enables NLI models to be successfully used for this task by segmenting documents into sentence units and aggregating scores between pairs of sentences. On our newly introduced benchmark called SummaC (Summary Consistency) consisting of six large inconsistency detection datasets, SummaCConv obtains state-of-the-art results with a balanced accuracy of 74.4%, a 5% point improvement compared to prior work. We make the models and datasets available: https://github.com/tingofurro/summac

Philippe Laban, Luke Dai, Lucas Bandarkar et al.

The Shuffle Test is the most common task to evaluate whether NLP models can measure coherence in text. Most recent work uses direct supervision on the task; we show that by simply finetuning a RoBERTa model, we can achieve a near perfect accuracy of 97.8%, a state-of-the-art. We argue that this outstanding performance is unlikely to lead to a good model of text coherence, and suggest that the Shuffle Test should be approached in a Zero-Shot setting: models should be evaluated without being trained on the task itself. We evaluate common models in this setting, such as Generative and Bi-directional Transformers, and find that larger architectures achieve high-performance out-of-the-box. Finally, we suggest the k-Block Shuffle Test, a modification of the original by increasing the size of blocks shuffled. Even though human reader performance remains high (around 95% accuracy), model performance drops from 94% to 78% as block size increases, creating a conceptually simple challenge to benchmark NLP models. Code available: https://github.com/tingofurro/shuffle_test/

Philippe Laban, Tobias Schnabel, Paul Bennett et al.

This work presents Keep it Simple (KiS), a new approach to unsupervised text simplification which learns to balance a reward across three properties: fluency, salience and simplicity. We train the model with a novel algorithm to optimize the reward (k-SCST), in which the model proposes several candidate simplifications, computes each candidate's reward, and encourages candidates that outperform the mean reward. Finally, we propose a realistic text comprehension task as an evaluation method for text simplification. When tested on the English news domain, the KiS model outperforms strong supervised baselines by more than 4 SARI points, and can help people complete a comprehension task an average of 18% faster while retaining accuracy, when compared to the original text. Code available: https://github.com/tingofurro/keep_it_simple

Philippe Laban, Hiroaki Hayashi, Yingbo Zhou et al. · microsoft-research

Large Language Models (LLMs) are conversational interfaces. As such, LLMs have the potential to assist their users not only when they can fully specify the task at hand, but also to help them define, explore, and refine what they need through multi-turn conversational exchange. Although analysis of LLM conversation logs has confirmed that underspecification occurs frequently in user instructions, LLM evaluation has predominantly focused on the single-turn, fully-specified instruction setting. In this work, we perform large-scale simulation experiments to compare LLM performance in single- and multi-turn settings. Our experiments confirm that all the top open- and closed-weight LLMs we test exhibit significantly lower performance in multi-turn conversations than single-turn, with an average drop of 39% across six generation tasks. Analysis of 200,000+ simulated conversations decomposes the performance degradation into two components: a minor loss in aptitude and a significant increase in unreliability. We find that LLMs often make assumptions in early turns and prematurely attempt to generate final solutions, on which they overly rely. In simpler terms, we discover that *when LLMs take a wrong turn in a conversation, they get lost and do not recover*.

Shuhaib Mehri, Philippe Laban, Sumuk Shashidhar et al.

As user simulators are increasingly used for interactive training and evaluation of AI assistants, it is essential that they represent the diverse behaviors of real users. While existing works train user simulators to generate human-like responses, whether they capture the broad and heterogeneous distribution of real user behaviors remains an open question. In this work, we introduce a method to measure the distributional gap between real and simulated user behaviors, validated through a human study and ablations. Given a dataset of real and simulated conversations, our method extracts representations of user behavior from each conversation, quantizes them into discrete distributions via clustering, then computes divergence metrics. We provide the first systematic evaluation of 24 LLM-based user simulators on coding and writing tasks, and reveal a large distributional gap from real users that varies across model families, scales, and behavioral facets. Pairwise comparisons show that most simulators behave similarly, while a few stand apart. Combining behaviorally complementary simulators brings the resulting distribution closer to real users compared to either simulator on its own. Finally, a TF-IDF analysis of the clusters surfaces interpretable patterns of behaviors that simulators capture, miss, and hallucinate.

Kung-Hsiang Huang, Akshara Prabhakar, Sidharth Dhawan et al. · microsoft-research, princeton

Customer Relationship Management (CRM) systems are vital for modern enterprises, providing a foundation for managing customer interactions and data. Integrating AI agents into CRM systems can automate routine processes and enhance personalized service. However, deploying and evaluating these agents is challenging due to the lack of realistic benchmarks that reflect the complexity of real-world CRM tasks. To address this issue, we introduce CRMArena, a novel benchmark designed to evaluate AI agents on realistic tasks grounded in professional work environments. Following guidance from CRM experts and industry best practices, we designed CRMArena with nine customer service tasks distributed across three personas: service agent, analyst, and manager. The benchmark includes 16 commonly used industrial objects (e.g., account, order, knowledge article, case) with high interconnectivity, along with latent variables (e.g., complaint habits, policy violations) to simulate realistic data distributions. Experimental results reveal that state-of-the-art LLM agents succeed in less than 40% of the tasks with ReAct prompting, and less than 55% even with function-calling abilities. Our findings highlight the need for enhanced agent capabilities in function-calling and rule-following to be deployed in real-world work environments. CRMArena is an open challenge to the community: systems that can reliably complete tasks showcase direct business value in a popular work environment.

Pranav Narayanan Venkit, Philippe Laban, Yilun Zhou et al. · microsoft-research

Large Language Model (LLM)-based applications are graduating from research prototypes to products serving millions of users, influencing how people write and consume information. A prominent example is the appearance of Answer Engines: LLM-based generative search engines supplanting traditional search engines. Answer engines not only retrieve relevant sources to a user query but synthesize answer summaries that cite the sources. To understand these systems' limitations, we first conducted a study with 21 participants, evaluating interactions with answer vs. traditional search engines and identifying 16 answer engine limitations. From these insights, we propose 16 answer engine design recommendations, linked to 8 metrics. An automated evaluation implementing our metrics on three popular engines (You.com, Perplexity.ai, BingChat) quantifies common limitations (e.g., frequent hallucination, inaccurate citation) and unique features (e.g., variation in answer confidence), with results mirroring user study insights. We release our Answer Engine Evaluation benchmark (AEE) to facilitate transparent evaluation of LLM-based applications.

Fan Yin, Philippe Laban, Xiangyu Peng et al. · microsoft-research

Malicious content generated by large language models (LLMs) can pose varying degrees of harm. Although existing LLM-based moderators can detect harmful content, they struggle to assess risk levels and may miss lower-risk outputs. Accurate risk assessment allows platforms with different safety thresholds to tailor content filtering and rejection. In this paper, we introduce per-topic severity rubrics for 11 harmful topics and build BingoGuard, an LLM-based moderation system designed to predict both binary safety labels and severity levels. To address the lack of annotations on levels of severity, we propose a scalable generate-then-filter framework that first generates responses across different severity levels and then filters out low-quality responses. Using this framework, we create BingoGuardTrain, a training dataset with 54,897 examples covering a variety of topics, response severity, styles, and BingoGuardTest, a test set with 988 examples explicitly labeled based on our severity rubrics that enables fine-grained analysis on model behaviors on different severity levels. Our BingoGuard-8B, trained on BingoGuardTrain, achieves the state-of-the-art performance on several moderation benchmarks, including WildGuardTest and HarmBench, as well as BingoGuardTest, outperforming best public models, WildGuard, by 4.3\%. Our analysis demonstrates that incorporating severity levels into training significantly enhances detection performance and enables the model to effectively gauge the severity of harmful responses.

Tuhin Chakrabarty, Philippe Laban, Chien-Sheng Wu · microsoft-research

AI-generated text is proliferating across domains, from creative writing and journalism to marketing content and scientific articles. Models can follow user-provided instructions to generate coherent and grammatically correct outputs but in this work, we study a more fundamental question: how do we evaluate and improve the writing quality of AI-generated text? Writing quality assessment has received less attention from the community, in part because it is fundamentally subjective and requires expertise. We first introduce the Writing Quality Benchmark (WQ) by consolidating five writing-preference datasets into 4,729 writing quality judgments. Our experiments show that most of the competitive baselines, including state-of-the-art LLMs that excel at reasoning tasks, barely outperform random baselines on WQ. We then train specialized Writing Quality Reward Models (WQRM) of various sizes for writing quality assessment that demonstrate strong generalization on four out-of-distribution test sets and 74% accuracy on the WQ benchmark. To further show WQRM's practical benefits during inference, we leverage additional test-time compute to generate and rank multiple candidate revisions, allowing us to select higher-quality outputs from an initial draft. Human evaluation with 9 experienced writers confirm that WQRM-based selection produces writing samples preferred by experts 66% overall, and 72.2% when the reward gap is larger than 1 point. We release our datasets and models to encourage community engagement with writing quality assessment and development of AI writing systems better aligned with human preferences.

Kung-Hsiang Huang, Can Qin, Haoyi Qiu et al. · microsoft-research

Vision Language Models (VLMs) have achieved remarkable progress in multimodal tasks, yet they often struggle with visual arithmetic, seemingly simple capabilities like object counting or length comparison, which are essential for relevant complex tasks like chart understanding and geometric reasoning. In this work, we first investigate the root causes of this deficiency through a suite of probing tasks focusing on basic visual arithmetic. Our analysis reveals that while pre-trained vision encoders typically capture sufficient information, the text decoder often fails to decode it correctly for arithmetic reasoning. To address this, we propose CogAlign, a novel post-training strategy inspired by Piaget's theory of cognitive development. CogAlign trains VLMs to recognize invariant properties under visual transformations. We demonstrate that this approach significantly improves the performance of three diverse VLMs on our proposed probing tasks. Furthermore, CogAlign enhances performance by an average of 4.6% on CHOCOLATE and 2.9% on MATH-VISION, outperforming or matching supervised fine-tuning methods while requiring only 60% less training data. These results highlight the effectiveness and generalizability of CogAlign in improving fundamental visual arithmetic capabilities and their transfer to downstream tasks.

Kaige Xie, Philippe Laban, Prafulla Kumar Choubey et al. · microsoft-research

Evaluating retrieval-augmented generation (RAG) systems remains challenging, particularly for open-ended questions that lack definitive answers and require coverage of multiple sub-topics. In this paper, we introduce a novel evaluation framework based on sub-question coverage, which measures how well a RAG system addresses different facets of a question. We propose decomposing questions into sub-questions and classifying them into three types -- core, background, and follow-up -- to reflect their roles and importance. Using this categorization, we introduce a fine-grained evaluation protocol that provides insights into the retrieval and generation characteristics of RAG systems, including three commercial generative answer engines: You.com, Perplexity AI, and Bing Chat. Interestingly, we find that while all answer engines cover core sub-questions more often than background or follow-up ones, they still miss around 50% of core sub-questions, revealing clear opportunities for improvement. Further, sub-question coverage metrics prove effective for ranking responses, achieving 82% accuracy compared to human preference annotations. Lastly, we also demonstrate that leveraging core sub-questions enhances both retrieval and answer generation in a RAG system, resulting in a 74% win rate over the baseline that lacks sub-questions.

Manya Wadhwa, Zayne Sprague, Chaitanya Malaviya et al.

Evaluation of language model outputs on structured writing tasks is typically conducted with a number of desirable criteria presented to human evaluators or large language models (LLMs). For instance, on a prompt like "Help me draft an academic talk on coffee intake vs research productivity", a model response may be evaluated for criteria like accuracy and coherence. However, high-quality responses should do more than just satisfy basic task requirements. An effective response to this query should include quintessential features of an academic talk, such as a compelling opening, clear research questions, and a takeaway. To help identify these implicit criteria, we introduce EvalAgent, a novel framework designed to automatically uncover nuanced and task-specific criteria. EvalAgent first mines expert-authored online guidance. It then uses this evidence to propose diverse, long-tail evaluation criteria that are grounded in reliable external sources. Our experiments demonstrate that the grounded criteria produced by EvalAgent are often implicit (not directly stated in the user's prompt), yet specific (high degree of lexical precision). Further, EvalAgent criteria are often not satisfied by initial responses but they are actionable, such that responses can be refined to satisfy them. Finally, we show that combining LLM-generated and EvalAgent criteria uncovers more human-valued criteria than using LLMs alone.

Peiqi Sui, Juan Diego Rodriguez, Philippe Laban et al.

Each year, tens of millions of essays are written and graded in college-level English courses. Students are asked to analyze literary and cultural texts through a process known as close reading, in which they gather textual details to formulate evidence-based arguments. Despite being viewed as a basis for critical thinking and widely adopted as a required element of university coursework, close reading has never been evaluated on large language models (LLMs), and multi-discipline benchmarks like MMLU do not include literature as a subject. To fill this gap, we present KRISTEVA, the first close reading benchmark for evaluating interpretive reasoning, consisting of 1331 multiple-choice questions adapted from classroom data. With KRISTEVA, we propose three progressively more difficult sets of tasks to approximate different elements of the close reading process, which we use to test how well LLMs may seem to understand and reason about literary works: 1) extracting stylistic features, 2) retrieving relevant contextual information from parametric knowledge, and 3) multi-hop reasoning between style and external contexts. Our baseline results find that, while state-of-the-art LLMs possess some college-level close reading competency (accuracy 49.7% - 69.7%), their performances still trail those of experienced human evaluators on 10 out of our 11 tasks.

Onkar Thorat, Philippe Laban, Chien-Sheng Wu · microsoft-research

Detecting factual inconsistencies in summarization is critical, yet existing benchmarks lack the necessary challenge and interpretability for robust evaluation. In this paper, we introduce SummExecEdit, a novel pipeline and benchmark leveraging executable edits to assess models on their ability to both detect factual errors and provide accurate explanations. The top-performing model, Claude3-Opus, achieves a joint detection and explanation score of only 0.49 in our benchmark, with individual scores of 0.67 for detection and 0.73 for explanation. We conduct detailed evaluations to assess the current state of models in this field and find that more than half of the 20+ LLMs in our study struggle with over 30% of the SummExecEdit benchmark. Additionally, we identify four primary types of explanation errors, with 45.4% of them involving a focus on completely unrelated parts of the summary.

Tarek Naous, Philippe Laban, Wei Xu et al. · microsoft-research

Conversations with LMs involve two participants: a human user leading the conversation, and an LM assistant responding to the user's request. To satisfy this specific role, LMs are post-trained to be helpful assistants -- optimized to produce exhaustive and well-structured responses, free of ambiguity and grammar errors. User utterances, on the other hand, are rarely perfected, with each user phrasing requests in unique ways, sometimes putting in partial effort at each turn and refining on the fly. To evaluate LM performance in realistic settings, prior work simulated users in multi-turn conversations, often prompting an LLM originally trained to be a helpful assistant to act as a user. However, we show that assistant LMs make for poor user simulators, with the surprising finding that better assistants yield worse simulators. Instead, we introduce purpose-built User Language Models (User LMs) - models post-trained to simulate human users in multi-turn conversations. Through various evaluations, we show how User LMs align better with human behavior and achieve better simulation robustness than existing simulation methods. When leveraging User LMs to simulate coding and math conversations, the performance of a strong assistant (GPT-4o) drops from 74.6% to 57.4%, confirming that more realistic simulation environments lead to assistant struggles as they fail to cope with the nuances of users in multi-turn setups.

Hita Kambhamettu, Alyssa Hwang, Philippe Laban et al. · microsoft-research

AI question answering systems increasingly generate responses with attributions to sources. However, the task of verifying the actual content of these attributions is in most cases impractical. In this paper, we present attribution gradients as a solution. Attribution gradients provide integrated, incremental affordances for diving into an attributed passage. A user can decompose a sentence of an answer into its claims. For each claim, the user can view supporting and contradictory excerpts mined from sources. Those excerpts serve as clickable conduits into the source (in our application, scientific papers). When evidence itself contains more citations, the UI unpacks the evidence into excerpts from the cited sources. These features of attribution gradients facilitate concurrent interconnections among answer, claim, excerpt, and context. In a usability study, we observed greater engagement with sources and richer revision in a task where participants revised an attributed AI answer with attribution gradients and a baseline.

Pranav Narayanan Venkit, Philippe Laban, Yilun Zhou et al. · microsoft-research

Generative search engines and deep research LLM agents promise trustworthy, source-grounded synthesis, yet users regularly encounter overconfidence, weak sourcing, and confusing citation practices. We introduce DeepTRACE, a novel sociotechnically grounded audit framework that turns prior community-identified failure cases into eight measurable dimensions spanning answer text, sources, and citations. DeepTRACE uses statement-level analysis (decomposition, confidence scoring) and builds citation and factual-support matrices to audit how systems reason with and attribute evidence end-to-end. Using automated extraction pipelines for popular public models (e.g., GPT-4.5/5, You.com, Perplexity, Copilot/Bing, Gemini) and an LLM-judge with validated agreement to human raters, we evaluate both web-search engines and deep-research configurations. Our findings show that generative search engines and deep research agents frequently produce one-sided, highly confident responses on debate queries and include large fractions of statements unsupported by their own listed sources. Deep-research configurations reduce overconfidence and can attain high citation thoroughness, but they remain highly one-sided on debate queries and still exhibit large fractions of unsupported statements, with citation accuracy ranging from 40--80% across systems.

Jiho Kim, Philippe Laban, Xiang 'Anthony' Chen et al.

Writing well requires not only expressing ideas but also refining them through revision, a process facilitated by reflection. Prior research suggests that feedback delivered through dialogues, such as those in writing center tutoring sessions, can help writers reflect more thoughtfully on their work compared to static feedback. Recent advancements in multi-modal large language models (LLMs) now offer new possibilities for supporting interactive and expressive voice-based reflection in writing. In particular, we propose that LLM-generated static feedback can be repurposed as conversation starters, allowing writers to seek clarification, request examples, and ask follow-up questions, thereby fostering deeper reflection on their writing. We argue that voice-based interaction can naturally facilitate this conversational exchange, encouraging writers' engagement with higher-order concerns, facilitating iterative refinement of their reflections, and reduce cognitive load compared to text-based interactions. To investigate these effects, we propose a formative study exploring how text vs. voice input influence writers' reflection and subsequent revisions. Findings from this study will inform the design of intelligent and interactive writing tools, offering insights into how voice-based interactions with LLM-powered conversational agents can support reflection and revision.

Philippe Laban, Jesse Vig, Wojciech Kryscinski et al.

Text simplification research has mostly focused on sentence-level simplification, even though many desirable edits - such as adding relevant background information or reordering content - may require document-level context. Prior work has also predominantly framed simplification as a single-step, input-to-output task, only implicitly modeling the fine-grained, span-level edits that elucidate the simplification process. To address both gaps, we introduce the SWiPE dataset, which reconstructs the document-level editing process from English Wikipedia (EW) articles to paired Simple Wikipedia (SEW) articles. In contrast to prior work, SWiPE leverages the entire revision history when pairing pages in order to better identify simplification edits. We work with Wikipedia editors to annotate 5,000 EW-SEW document pairs, labeling more than 40,000 edits with proposed 19 categories. To scale our efforts, we propose several models to automatically label edits, achieving an F-1 score of up to 70.6, indicating that this is a tractable but challenging NLU task. Finally, we categorize the edits produced by several simplification models and find that SWiPE-trained models generate more complex edits while reducing unwanted edits.

Philippe Laban, Wojciech Kryściński, Divyansh Agarwal et al.

With the recent appearance of LLMs in practical settings, having methods that can effectively detect factual inconsistencies is crucial to reduce the propagation of misinformation and improve trust in model outputs. When testing on existing factual consistency benchmarks, we find that a few large language models (LLMs) perform competitively on classification benchmarks for factual inconsistency detection compared to traditional non-LLM methods. However, a closer analysis reveals that most LLMs fail on more complex formulations of the task and exposes issues with existing evaluation benchmarks, affecting evaluation precision. To address this, we propose a new protocol for inconsistency detection benchmark creation and implement it in a 10-domain benchmark called SummEdits. This new benchmark is 20 times more cost-effective per sample than previous benchmarks and highly reproducible, as we estimate inter-annotator agreement at about 0.9. Most LLMs struggle on SummEdits, with performance close to random chance. The best-performing model, GPT-4, is still 8\% below estimated human performance, highlighting the gaps in LLMs' ability to reason about facts and detect inconsistencies when they occur.

Philippe Laban, Elicia Ye, Srujay Korlakunta et al.

News podcasts are a popular medium to stay informed and dive deep into news topics. Today, most podcasts are handcrafted by professionals. In this work, we advance the state-of-the-art in automatically generated podcasts, making use of recent advances in natural language processing and text-to-speech technology. We present NewsPod, an automatically generated, interactive news podcast. The podcast is divided into segments, each centered on a news event, with each segment structured as a Question and Answer conversation, whose goal is to engage the listener. A key aspect of the design is the use of distinct voices for each role (questioner, responder), to better simulate a conversation. Another novel aspect of NewsPod allows listeners to interact with the podcast by asking their own questions and receiving automatically generated answers. We validate the soundness of this system design through two usability studies, focused on evaluating the narrative style and interactions with the podcast, respectively. We find that NewsPod is preferred over a baseline by participants, with 80% claiming they would use the system in the future.

Lidiya Murakhovs'ka, Chien-Sheng Wu, Philippe Laban et al.

Asking good questions is an essential ability for both human and machine intelligence. However, existing neural question generation approaches mainly focus on the short factoid type of answers. In this paper, we propose a neural question generator, MixQG, to bridge this gap. We combine 9 question answering datasets with diverse answer types, including yes/no, multiple-choice, extractive, and abstractive answers, to train a single generative model. We show with empirical results that our model outperforms existing work in both seen and unseen domains and can generate questions with different cognitive levels when conditioned on different answer types. Our code is released and well-integrated with the Huggingface library to facilitate various downstream applications.

Philippe Laban, John Canny, Marti A. Hearst

This work describes an automatic news chatbot that draws content from a diverse set of news articles and creates conversations with a user about the news. Key components of the system include the automatic organization of news articles into topical chatrooms, integration of automatically generated questions into the conversation, and a novel method for choosing which questions to present which avoids repetitive suggestions. We describe the algorithmic framework and present the results of a usability study that shows that news readers using the system successfully engage in multi-turn conversations about specific news stories.

Philippe Laban, Lucas Bandarkar, Marti A. Hearst

Recent progress in Natural Language Understanding (NLU) has seen the latest models outperform human performance on many standard tasks. These impressive results have led the community to introspect on dataset limitations, and iterate on more nuanced challenges. In this paper, we introduce the task of HeadLine Grouping (HLG) and a corresponding dataset (HLGD) consisting of 20,056 pairs of news headlines, each labeled with a binary judgement as to whether the pair belongs within the same group. On HLGD, human annotators achieve high performance of around 0.9 F-1, while current state-of-the art Transformer models only reach 0.75 F-1, opening the path for further improvements. We further propose a novel unsupervised Headline Generator Swap model for the task of HeadLine Grouping that achieves within 3 F-1 of the best supervised model. Finally, we analyze high-performing models with consistency tests, and find that models are not consistent in their predictions, revealing modeling limits of current architectures.

Philippe Laban, Andrew Hsi, John Canny et al.

This work presents a new approach to unsupervised abstractive summarization based on maximizing a combination of coverage and fluency for a given length constraint. It introduces a novel method that encourages the inclusion of key terms from the original document into the summary: key terms are masked out of the original document and must be filled in by a coverage model using the current generated summary. A novel unsupervised training procedure leverages this coverage model along with a fluency model to generate and score summaries. When tested on popular news summarization datasets, the method outperforms previous unsupervised methods by more than 2 R-1 points, and approaches results of competitive supervised methods. Our model attains higher levels of abstraction with copied passages roughly two times shorter than prior work, and learns to compress and merge sentences without supervision.