Users of feature-rich tools like Excel often miss more efficient workflows, repeating tedious steps and making avoidable errors. Current approaches to helping them require either manual prompting, which is effortful for users, or automated logging, which is limiting for developers. We present InvisibleMentor, a system inspired by over-the-shoulder learning: it observes what users do, then shows them how to do it better. To do this, InvisibleMentor analyzes screen recordings with a vision-language model to reconstruct actions and context, then uses a large language model to generate vision-grounded task reflection, structured suggestions grounded in observed behavior. In a user study, participants found InvisibleMentor’s suggestions more clear, more relevant, and more useful than those from a prompt-based assistant, demonstrating that AI can do more than automate away work—it can help users master it.

Yue Yang, Ajay Patel, Matt Deitke, Tanmay Gupta, Luca Weihs, Andrew Head, Mark Yatskar, Chris Callison-Burch, Ranjay Krishna, Aniruddha Kembhavi, Christopher Clark. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Today’s most advanced vision-language models (VLMs) remain proprietary. The strongest open-weight models rely heavily on synthetic data from proprietary VLMs to achieve good performance, effectively distilling these closed VLMs into open ones. As a result, the community has been missing foundational knowledge about how to build performant VLMs from scratch. We present Molmo, a new family of VLMs that are state-of-the-art in their class of openness. Our key contribution is a collection of new datasets called PixMo, including a dataset of highly detailed image captions for pre-training, a free-form image Q&A dataset for fine-tuning, and an innovative 2D pointing dataset, all collected without the use of external VLMs. The success of our approach relies on careful modeling choices, a well- tuned training pipeline, and, most critically, the quality of our newly collected datasets. Our best-in-class 72B model not only outperforms others in the class of open weight and data models, but also outperforms larger proprietary models including Claude 3.5 Sonnet, and Gemini 1.5 Pro and Flash, second only to GPT-4o based on both academic benchmarks and on a large human evaluation. Our model weights, new datasets, and source code are available at https://molmo.allenai.org/blog.

Interactive theorem provers, or proof assistants, are important tools across many areas of computer science and mathematics, but even experts find them challenging to use effectively. To improve their design, we need a deeper, user-centric understanding of proof assistant usage. We present the results of an observation study of proof assistant users. We use contextual inquiry methodology, observing 30 participants doing their everyday work in Rocq and Lean. We qualitatively analyze their experiences to surface four observations: that proof writers iterate on their proofs by reacting to and incorporating feedback from the proof assistant; that proof progress often involves challenging conversations with the proof assistant; that proofs are constructed in consultation with a wide array of external resources; and that proof writers are guided by design considerations that go beyond “getting to QED.” Our documentation of these themes clarifies what proof assistant usage looks like currently and identifies potential opportunities that researchers should consider when working to improve the usability of proof assistants.

Medical progress notes have recently become available to patients at an unprecedented scale. Progress notes offer patients insight into their care that they cannot find elsewhere. That said, reading a note requires patients to contend with the language, unspoken assumptions, and clutter common to clinical documentation. As the health system reinvents many of its interfaces to incorporate AI assistance, this paper examines what intelligent interfaces could do to help patients read their progress notes. In a qualitative study, we examine the needs of patients as they read a progress note. We then formulate a vision for the explainable note, an augmented progress note that provides support for directing attention, phrase-level understanding, and tracing lines of reasoning. This vision manifests in a set of patient-inspired opportunities for advancing intelligent interfaces for writing and reading progress notes.

Programming assistants have reshaped the experience of programming into one where programmers spend less time writing and more time critically examining code. In this paper, we explore how programming assistants can be extended to accelerate the inspection of generated code. We introduce an extension to the programming assistant called Ivie, or instantly visible in-situ explanations. When using Ivie, a programmer’s generated code is instantly accompanied by explanations positioned just adjacent to the code. Our design was optimized for low-cost invocation and dismissal. Explanations are compact and informative. They describe meaningful expressions, from individual variables to entire blocks of code. We present an implementation of Ivie that forks VS Code, applying a modern LLM for timely segmentation and explanation of generated code. In a lab study, we compared Ivie to a contemporary baseline tool for code understanding. Ivie improved understanding of generated code, and was received by programmers as a highly useful, low distraction complement to the programming assistant.

Property-based testing (PBT) is a testing methodology where users write executable formal specifications of software components and an automated harness checks these specifications against many automatically generated inputs. From its roots in the QuickCheck library in Haskell, PBT has made significant inroads in mainstream languages and industrial practice at companies such as Amazon, Volvo, and Stripe. As PBT extends its reach, it is important to understand how developers are using it in practice, where they see its strengths and weaknesses, and what innovations are needed to make it more effective.

Software developers increasingly rely on automated methods to assess the correctness of their code. One such method is property-based testing (PBT), wherein a test harness generates hundreds or thousands of inputs and checks the outputs of the program on those inputs using parametric properties. Though powerful, PBT induces a sizable gulf of evaluation: developers need to put in nontrivial effort to understand how well the different test inputs exercise the software under test. To bridge this gulf, we propose Tyche, a user interface that supports sensemaking around the effectiveness of property-based tests. Guided by a formative design exploration, our design of Tyche supports developers with interactive, configurable views of test behavior with tight integrations into modern developer testing workflow. These views help developers explore global testing behavior and individual test inputs alike. To accelerate the development of powerful, interactive PBT tools, we define a standard for PBT test reporting and integrate it with a widely used PBT library. A self-guided online usability study revealed that Tyche’s visualizations help developers to more accurately assess software testing effectiveness.

Scholars need to keep up with an exponentially increasing flood of scientific papers. To aid this challenge, we introduce Scim , a novel intelligent interface that helps scholars skim papers to rapidly review and gain a cursory understanding of its contents. Scim supports the skimming process by highlighting salient content within a paper, directing a scholar’s attention. These automatically-extracted highlights are faceted by content type, evenly-distributed across a paper, and have a density configurable by scholars. We evaluate Scim with an in-lab usability study and a longitudinal diary study, revealing how its highlights facilitate the more efficient construction of a conceptualization of a paper. Finally, we describe the process of scaling highlights from their conception within Scim , a research prototype, to production on over 521,000 papers within the Semantic Reader, a publicly-available augmented reading interface for scientific papers. We conclude by discussing design considerations and tensions for the design of future skimming tools with augmented intelligence.

Voice assistants have sharply risen in popularity in recent years, but their use has been limited mostly to simple applications like music, hands-free search, or control of internet-of-things devices. What would it take for voice assistants to guide people through more complex tasks? In our work, we study the limitations of the dominant approach voice assistants take to complex task guidance: reading aloud written instructions. Using recipes as an example, we observe twelve participants cook at home with a state-of-the-art voice assistant. We learn that the current approach leads to nine challenges, including obscuring the bigger picture, overwhelming users with too much information, and failing to communicate affordances. Instructions delivered by a voice assistant are especially difficult because they cannot be skimmed as easily as written instructions. Alexa in particular did not surface crucial details to the user or answer questions well. We draw on our observations to propose eight ways in which voice assistants can “rewrite the script”—summarizing, signposting, splitting, elaborating, volunteering, reordering, redistributing, and visualizing—to transform written sources into forms that are readily communicated through spoken conversation. We conclude with a vision of how modern advancements in natural language processing can be leveraged for intelligent agents to guide users effectively through complex tasks.

When reading a scholarly article, inline citations help researchers contextualize the current article and discover relevant prior work. However, it can be challenging to prioritize and make sense of the hundreds of citations encountered during literature reviews. This paper introduces CiteSee, a paper reading tool that leverages a user’s publishing, reading, and saving activities to provide personalized visual augmentations and context around citations. First, CiteSee connects the current paper to familiar contexts by surfacing known citations a user had cited or opened. Second, CiteSee helps users prioritize their exploration by highlighting relevant but unknown citations based on saving and reading history. We conducted a lab study that suggests CiteSee is significantly more effective for paper discovery than three baselines. A field deployment study shows CiteSee helps participants keep track of their explorations and leads to better situational awareness and increased paper discovery via inline citation when conducting real-world literature reviews.

Multi-stage programming tutorials are key learning resources for programmers, using progressive incremental steps to teach them how to build larger software systems. A good multi-stage tutorial describes the code clearly, explains the rationale and code changes for each step, and allows readers to experiment as they work through the tutorial. In practice, it is time-consuming for authors to create tutorials with these attributes. In this paper, we introduce Colaroid, an interactive authoring tool for creating high quality multi-stage tutorials. Colaroid tutorials are augmented computational notebooks, where snippets and outputs represent a snapshot of a project, with source code differences highlighted, complete source code context for each snippet, and the ability to load and tinker with any stage of the project in a linked IDE. In two laboratory studies, we found Colaroid makes it easy to create multi-stage tutorials, while offering advantages to readers compared to video and web-based tutorials.

Scholars need to keep up with an exponentially increasing flood of scientific papers. To aid this challenge, we introduce Scim, a novel intelligent interface that helps experienced researchers skim – or rapidly review – a paper to attain a cursory understanding of its contents. Scim supports the skimming process by highlighting salient paper contents in order to direct a reader’s attention. The system’s highlights are faceted by content type, evenly distributed across a paper, and have a density configurable by readers at both the global and local level. We evaluate Scim with both an in-lab usability study and a longitudinal diary study, revealing how its highlights facilitate the more efficient construction of a conceptualization of a paper. We conclude by discussing design considerations and tensions for the design of future intelligent skimming tools.

As interest grows in learning math concepts in fields like data science and machine learning, it is becoming more important to help broad audiences engage with math notation. In this paper, we explore how authoring tools can help authors better style and label formulas to support their readability. We introduce a markup language for augmenting formulas called FFL, or “Formula Formatting Language,” which aims to lower the threshold to stylize and diagram formulas. The language is designed to be concise, writable, readable, and integrable into web-based document authoring environments. It was developed with an accompanying runtime that supports live application of augmentations to formulas. Our lab study shows that FFL improves the speed and ease of editing augmentation markup, and the readability of augmentation markup compared to baseline LaTeX tools. These results clarify the role tooling can play in supporting the explanation of math notation.

When seeking information not covered in patient-friendly documents, healthcare consumers may turn to the research literature. Reading medical papers, however, can be a challenging experience. To improve access to medical papers, we explore four features enabled by natural language processing: definitions of unfamiliar terms, in-situ plain language section summaries, a collection of key questions that guides readers to answering passages, and plain language summaries of those passages. We embody these features into a prototype system, Paper Plain . We evaluate Paper Plain , finding that participants who used the prototype system had an easier time reading research papers without a loss in paper comprehension compared to those who used a typical PDF reader. Altogether, the study results suggest that guiding readers to relevant passages and providing plain language summaries alongside the original paper content can make reading medical papers easier and give readers more confidence to approach these papers.

With the increasing growth and impact of machine learning and other math-intensive fields, it is more important than ever to broaden access to mathematical notation. Can new visual and interactive displays help a wider readership successfully engage with notation? This paper provides the first detailed qualitative analysis of math augmentation—the practice of embellishing notation with novel visual design patterns to improve its readability. We present two qualitative studies of the practice of math augmentation. First is an analysis of 1.1k augmentations to 281 formulas in 47 blogs, textbooks, and other documents containing mathematical expressions. Second is an interview study with 12 authors who had previously designed custom math augmentations (“maugs”). This paper contributes a comprehensive inventory of the kinds of maugs that appear in math documents, and a detailed account of how authors’ tools ought to be redesigned to support efficient creation of math augmentations. These studies open a critical new design space for HCI researchers and interface designers.

The ever-increasing pace of scientific publication necessitates methods for quickly identifying relevant papers. While neural recommenders trained on user interests can help, they still result in long, monotonous lists of suggested papers. To improve the discovery experience we introduce multiple new methods for augmenting recommendations with textual relevance messages that highlight knowledge-graph connections between recommended papers and a user’s publication and interaction history. We explore associations mediated by author entities and those using citations alone. In a large-scale, real-world study, we show how our approach significantly increases engagement—and future engagement when mediated by authors—without introducing bias towards highly-cited authors. To expand message coverage for users with less publication or interaction history, we develop a novel method that highlights connections with proxy authors of interest to users and evaluate it in a controlled lab study. Finally, we synthesize design implications for future graph-based messages.

Computational notebooks are commonly used for iterative workflows, such as in exploratory data analysis. This process lends itself to the accumulation of old code and hidden state, making it hard for users to reason about the lineage of, e.g., plots depicting insights or trained machine learning models. One way to reason about code used to generate various notebook data artifacts is to compute a program slice , but traditional static approaches to slicing can be both inaccurate (failing to contain relevant code for artifacts) and conservative (containing unnecessary code for an artifacts). We present nbslicer, a dynamic slicer optimized for the notebook setting whose instrumentation for resolving dynamic data dependencies is both bolt-on (and therefore portable) and switchable (allowing it to be selectively disabled in order to reduce instrumentation overhead). We demonstrate Nbslicer's ability to construct small and accurate backward slices (i.e., historical cell dependencies) and forward slices (i.e., cells affected by the "rerun" of an earlier cell), thereby improving reproducibility in notebooks and enabling faster reactive re-execution, respectively. Comparing nbslicer with a static slicer on 374 real notebook sessions, we found that nbslicer filters out far more superfluous program statements while maintaining slice correctness, giving slices that are, on average, 66% and 54% smaller for backward and forward slices, respectively.

Despite the central importance of research papers to scientific progress, they can be difficult to read. Comprehension is often stymied when the information needed to understand a passage resides somewhere else—in another section, or in another paper. In this work, we envision how interfaces can bring definitions of technical terms and symbols to readers when and where they need them most. We introduce ScholarPhi, an augmented reading interface with four novel features: (1) tooltips that surface position-sensitive definitions from elsewhere in a paper, (2) a filter over the paper that “declutters” it to reveal how the term or symbol is used across the paper, (3) automatic equation diagrams that expose multiple definitions in parallel, and (4) an automatically generated glossary of important terms and symbols. A usability study showed that the tool helps researchers of all experience levels read papers. Furthermore, researchers were eager to have ScholarPhi’s definitions available to support their everyday reading.

Computational notebooks have emerged as the platform of choice for data science and analytical workflows, enabling rapid iteration and exploration. By keeping intermediate program state in memory and segmenting units of execution into so-called "cells", notebooks allow users to enjoy particularly tight feedback. However, as cells are added, removed, reordered, and rerun, this hidden intermediate state accumulates, making execution behavior difficult to reason about, and leading to errors and lack of reproducibility. We present nbsafety, a custom Jupyter kernel that uses runtime tracing and static analysis to automatically manage lineage associated with cell execution and global notebook state. nbsafety detects and prevents errors that users make during unaided notebook interactions, all while preserving the flexibility of existing notebook semantics. We evaluate nbsafety's ability to prevent erroneous interactions by replaying and analyzing 666 real notebook sessions. Of these, nbsafety identified 117 sessions with potential safety errors, and in the remaining 549 sessions, the cells that nbsafety identified as resolving safety issues were more than 7X more likely to be selected by users for re-execution compared to a random baseline, even though the users were not using nbsafety and were therefore not influenced by its suggestions.

Programming tutorials are a pervasive, versatile medium for teaching programming. In this paper, we report on the content and structure of programming tutorials, the pain points authors experience in writing them, and a design for a tool to help improve this process. An interview study with 12 experienced tutorial authors found that they construct documents by interleaving code snippets with text and illustrative outputs. It also revealed that authors must often keep related artifacts of source programs, snippets, and outputs consistent as a program evolves. A content analysis of 200 frequently-referenced tutorials on the web also found that most tutorials contain related artifacts—duplicate code and outputs generated from snippets—that an author would need to keep consistent with each other. To address these needs, we designed a tool called Torii with novel authoring capabilities. An in-lab study showed that tutorial authors can successfully use the tool for the unique affordances identified, and provides guidance for designing future tools for tutorial authoring.

Data analysts use computational notebooks to write code for analyzing and visualizing data. Notebooks help analysts iteratively write analysis code by letting them interleave code with output, and selectively execute cells. However, as analysis progresses, analysts leave behind old code and outputs, and overwrite important code, producing cluttered and inconsistent notebooks. This paper introduces code gathering tools, extensions to computational notebooks that help analysts find, clean, recover, and compare versions of code in cluttered, inconsistent notebooks. The tools archive all versions of code outputs, allowing analysts to review these versions and recover the subsets of code that produced them. These subsets can serve as succinct summaries of analysis activity or starting points for new analyses. In a qualitative usability study, 12 professional analysts found the tools useful for cleaning notebooks and writing analysis code, and discovered new ways to use them, like generating personal documentation and lightweight versioning.

Without good models and the right tools to interpret them, data scientists risk making decisions based on hidden biases, spurious correlations, and false generalizations. This has led to a rallying cry for model interpretability. Yet the concept of interpretability remains nebulous, such that researchers and tool designers lack actionable guidelines for how to incorporate interpretability into models and accompanying tools. Through an iterative design process with expert machine learning researchers and practitioners, we designed a visual analytics system, Gamut, to explore how interactive interfaces could better support model interpretation. Using Gamut as a probe, we investigated why and how professional data scientists interpret models, and how interface affordances can support data scientists in answering questions about model interpretability. Our investigation showed that interpretability is not a monolithic concept: data scientists have different reasons to interpret models and tailor explanations for specific audiences, often balancing competing concerns of simplicity and completeness. Participants also asked to use Gamut in their work, highlighting its potential to help data scientists understand their own data.

Programmers frequently learn from examples produced and shared by other programmers. However, it can be challenging and time-consuming to produce concise, working code examples. We conducted a formative study where 12 participants made examples based on their own code. This revealed a key hurdle: making meaningful simplifications without introducing errors. Based on this insight, we designed a mixed-initiative tool, CodeScoop, to help programmers extract executable, simplified code from existing code. CodeScoop enables programmers to "scoop" out a relevant subset of code. Techniques include selectively including control structures and recording an execution trace that allows authors to substitute literal values for code and variables. In a controlled study with 19 participants, CodeScoop helped programmers extract executable code examples with the intended behavior more easily than with a standard code editor.

Without usable and accurate documentation of how to use an API, developers can find themselves deterred from reusing relevant code. In C++, one place developers can find documentation is in a header file. When information is missing, they may look at the corresponding implementation code. To understand what's missing from C++ API documentation and the factors influencing whether it will be fixed, we conducted a mixed-methods study involving two experience sampling surveys with hundreds of developers at the moment they visited implementation code, interviews with 18 of those developers, and interviews with 8 API maintainers. In many cases, updating documentation may provide only limited value for developers, while requiring effort maintainers don't want to invest. We identify a set of questions maintainers and tool developers should consider when improving API-level documentation.

The rise in prevalence of Internet of Things (IoT) technologies has encouraged more people to prototype and build custom internet connected devices based on low power microcontrollers. While well-developed tools exist for debugging network communication for desktop and web applications, it can be difficult for developers of networked embedded systems to figure out why their network code is failing due to the limited output affordances of embedded devices. This paper presents WiFröst, a new approach for debugging these systems using instrumentation that spans from the device itself, to its communication API, to the wireless router and back-end server. WiFröst automatically collects this data, displays it in a web-based visualization, and highlights likely issues with an extensible suite of checks based on analysis of recorded execution traces.

In many developing country environments, it is difficult or impossible to obtain recent, reliable estimates of human development. Nationally representative household surveys, which are the standard instrument for determining development policy and priorities, are typically too expensive to collect with any regularity. Recently, however, researchers have shown the potential for remote sensing technologies to provide a possible solution to this data constraint. In particular, recent work indicates that satellite imagery can be processed with deep neural networks to accurately estimate the sub-regional distribution of wealth in sub-Saharan Africa.

In large introductory programming classes, teacher feedback on individual incorrect student submissions is often infeasible. Program synthesis techniques are capable of fixing student bugs and generating hints automatically, but they lack the deep domain knowledge of a teacher and can generate functionally correct but stylistically poor fixes. We introduce a mixed-initiative approach which combines teacher expertise with data-driven program synthesis techniques. We demonstrate our novel approach in two systems that use different interaction mechanisms. Our systems use program synthesis to learn bug-fixing code transformations and then cluster incorrect submissions by the transformations that correct them. The MistakeBrowser system learns transformations from examples of students fixing bugs in their own submissions. The FixPropagator system learns transformations from teachers fixing bugs in incorrect student submissions. Teachers can write feedback about a single submission or a cluster of submissions and propagate the feedback to all other submissions that can be fixed by the same transformation. Two studies suggest this approach helps teachers better understand student bugs and write reusable feedback that scales to a massive introductory programming classroom.

Recent advances in program synthesis offer means to automatically debug student submissions and generate personalized feedback in massive programming classrooms. When automatically generating feedback for programming assignments, a key challenge is designing pedagogically useful hints that are as effective as the manual feedback given by teachers. Through an analysis of teachers' hint-giving practices in 132 online Q&A posts, we establish three design guidelines that an effective feedback design should follow. Based on these guidelines, we develop a feedback system that leverages both program synthesis and visualization techniques. Our system compares the dynamic code execution of both incorrect and fixed code and highlights how the error leads to a difference in behavior and where the incorrect code trace diverges from the expected solution. Results from our study suggest that our system enables students to detect and fix bugs that are not caught by students using another existing visual debugging tool.

Developers choose open source packages from many alternatives. One increasingly important factor when choosing a package is its "social health", or a developer’s ability to get help on communication channels. We conduct a study to understand how developers learn about the social health of open source packages before using them. We offer preliminary results of the cues developers find.

We describe Lamello, an approach for creating tangible input components that recognize user interaction via passive acoustic sensing. Lamello employs comb-like structures with varying-length tines at interaction points (e.g., along slider paths). Moving a component generates tine strikes; a real-time audio processing pipeline analyzes the resultant sounds and emits high-level interaction events. Our main contributions are in the co-design of the tine structures, information encoding schemes, and audio analysis. We demonstrate 3D printed Lamello-powered buttons, sliders, and dials.

Programmers frequently turn to the web to solve problems and find example code. For the sake of brevity, the snippets in online instructions often gloss over the syntax of languages like CSS selectors and Unix commands. Programmers must compensate by consulting external documentation. In this paper, we propose language-specific routines called Tutorons that automatically generate context-relevant, on-demand micro-explanations of code. A Tutoron detects explainable code in a web page, parses it, and generates in-situ natural language explanations and demonstrations of code. We build Tutorons for CSS selectors, regular expressions, and the Unix command “wget”. We demonstrate techniques for generating natural language explanations through template instantiation, synthesizing code demonstrations by parse tree traversal, and building compound explanations of co-occurring options. Through a qualitative study, we show that Tutoron-generated explanations can reduce the need for reference documentation in code modification tasks.