Juho Kim

Seonghyeon Ye, Doyoung Kim, Sungdong Kim et al. · cmu

Evaluation of Large Language Models (LLMs) is challenging because instruction-following necessitates alignment with human values and the required set of skills varies depending on the instruction. However, previous studies have mainly focused on coarse-grained evaluation (i.e. overall preference-based evaluation), which limits interpretability since it does not consider the nature of user instructions that require instance-wise skill composition. In this paper, we introduce FLASK (Fine-grained Language Model Evaluation based on Alignment Skill Sets), a fine-grained evaluation protocol for both human-based and model-based evaluation which decomposes coarse-level scoring to a skill set-level scoring for each instruction. We experimentally observe that the fine-graininess of evaluation is crucial for attaining a holistic view of model performance and increasing the reliability of the evaluation. Using FLASK, we compare multiple open-source and proprietary LLMs and observe a high correlation between model-based and human-based evaluations. We publicly release the evaluation data and code implementation at https://github.com/kaistAI/FLASK.

Nayeon Lee, Chani Jung, Junho Myung et al.

Warning: this paper contains content that may be offensive or upsetting. Most hate speech datasets neglect the cultural diversity within a single language, resulting in a critical shortcoming in hate speech detection. To address this, we introduce CREHate, a CRoss-cultural English Hate speech dataset. To construct CREHate, we follow a two-step procedure: 1) cultural post collection and 2) cross-cultural annotation. We sample posts from the SBIC dataset, which predominantly represents North America, and collect posts from four geographically diverse English-speaking countries (Australia, United Kingdom, Singapore, and South Africa) using culturally hateful keywords we retrieve from our survey. Annotations are collected from the four countries plus the United States to establish representative labels for each country. Our analysis highlights statistically significant disparities across countries in hate speech annotations. Only 56.2% of the posts in CREHate achieve consensus among all countries, with the highest pairwise label difference rate of 26%. Qualitative analysis shows that label disagreement occurs mostly due to different interpretations of sarcasm and the personal bias of annotators on divisive topics. Lastly, we evaluate large language models (LLMs) under a zero-shot setting and show that current LLMs tend to show higher accuracies on Anglosphere country labels in CREHate. Our dataset and codes are available at: https://github.com/nlee0212/CREHate

Juho Kim

PokerKit is an open-source Python library designed to overcome the restrictions of existing poker game simulation and hand evaluation tools, which typically support only a handful of poker variants and lack flexibility in game state control. In contrast, PokerKit significantly expands this scope by supporting an extensive array of poker variants and it provides a flexible architecture for users to define their custom games. This paper details the design and implementation of PokerKit, including its intuitive programmatic API, multi-variant game support, and a unified hand evaluation suite across different hand types. The flexibility of PokerKit allows for applications in diverse areas, such as poker AI development, tool creation, and online poker casino implementation. PokerKit's reliability has been established through static type checking, extensive doctests, and unit tests, achieving 99% code coverage. The introduction of PokerKit represents a significant contribution to the field of computer poker, fostering future research and advanced AI development for a wide variety of poker games. The source code is available at https://github.com/uoftcprg/pokerkit

Tae Soo Kim, Yoonjoo Lee, Jamin Shin et al.

By simply composing prompts, developers can prototype novel generative applications with Large Language Models (LLMs). To refine prototypes into products, however, developers must iteratively revise prompts by evaluating outputs to diagnose weaknesses. Formative interviews (N=8) revealed that developers invest significant effort in manually evaluating outputs as they assess context-specific and subjective criteria. We present EvalLM, an interactive system for iteratively refining prompts by evaluating multiple outputs on user-defined criteria. By describing criteria in natural language, users can employ the system's LLM-based evaluator to get an overview of where prompts excel or fail, and improve these based on the evaluator's feedback. A comparative study (N=12) showed that EvalLM, when compared to manual evaluation, helped participants compose more diverse criteria, examine twice as many outputs, and reach satisfactory prompts with 59% fewer revisions. Beyond prompts, our work can be extended to augment model evaluation and alignment in specific application contexts.

DaEun Choi, Kihoon Son, Jaesang Yu et al.

While designers increasingly leverage Generative AI for divergent exploration, current interaction is optimized for convergent refinement, forcing users to specify fixed targets rather than open-ended search spaces. Based on a formative study (N=7), we define the anatomy of Divergent Intent, comprising property, direction, and range, and identified two critical barriers: the lack of mechanisms to explicitly shape the parametric boundaries of exploration and the difficulty of reusing successful search strategies. We present IdeaBlocks, where users can modularize divergent intents into Exploration Blocks. Users can reuse prior intents at multiple levels (block, path, and project) with options for literal or context-adaptive reuse. In our comparative study (N=12), participants using IdeaBlocks explored 2.13 times more images with 12.5% greater visual diversity than the baseline, demonstrating how structured intent expression and reuse support effective divergence. A three-day deployment study (N=6) further revealed how different reuse mechanisms allowed distinct creative strategies, offering design implications for future intent-aware creativity supports.

Jieun Han, Haneul Yoo, Junho Myung et al.

In the context of English as a Foreign Language (EFL) writing education, LLM-as-a-tutor can assist students by providing real-time feedback on their essays. However, challenges arise in assessing LLM-as-a-tutor due to differing standards between educational and general use cases. To bridge this gap, we integrate pedagogical principles to assess student-LLM interaction. First, we explore how LLMs can function as English tutors, providing effective essay feedback tailored to students. Second, we propose three metrics to evaluate LLM-as-a-tutor specifically designed for EFL writing education, emphasizing pedagogical aspects. In this process, EFL experts evaluate the feedback from LLM-as-a-tutor regarding quality and characteristics. On the other hand, EFL learners assess their learning outcomes from interaction with LLM-as-a-tutor. This approach lays the groundwork for developing LLMs-as-a-tutor tailored to the needs of EFL learners, advancing the effectiveness of writing education in this context.

Tae Soo Kim, Arghya Sarkar, Yoonjoo Lee et al.

Large language models (LLMs) can enhance writing by automating or supporting specific tasks in writers' workflows (e.g., paraphrasing, creating analogies). Leveraging this capability, a collection of interfaces have been developed that provide LLM-powered tools for specific writing tasks. However, these interfaces provide limited support for writers to create personal tools for their own unique tasks, and may not comprehensively fulfill a writer's needs -- requiring them to continuously switch between interfaces during writing. In this work, we envision LMCanvas, an interface that enables writers to create their own LLM-powered writing tools and arrange their personal writing environment by interacting with "blocks" in a canvas. In this interface, users can create text blocks to encapsulate writing and LLM prompts, model blocks for model parameter configurations, and connect these to create pipeline blocks that output generations. In this workshop paper, we discuss the design for LMCanvas and our plans to develop this concept.

Jeongeon Park, Bryan Min, Kihoon Son et al.

From deciding on a PhD program to buying a new camera, unfamiliar decisions--decisions without domain knowledge--are frequent and significant. The complexity and uncertainty of such decisions demand unique approaches to information seeking, understanding, and decision-making. Our formative study highlights that users want to start by discovering broad and relevant domain information evenly and simultaneously, quickly address emerging inquiries, and gain personalized standards to assess information found. We present ChoiceMates, an interactive multi-agent system designed to address these needs by enabling users to engage with a dynamic set of LLM agents each presenting a unique experience in the domain. Unlike existing multi-agent systems that automate tasks with agents, the user orchestrates agents to assist their decision-making process. Our user evaluation (n=12) shows that ChoiceMates enables a more confident, satisfactory decision-making with better situation understanding than web search, and higher decision quality and confidence than a commercial multi-agent framework. This work provides insights into designing a more controllable and collaborative multi-agent system.

Saelyne Yang, Jaesang Yu, Yi-Hao Peng et al.

Graphical User Interface (GUI) agents have the potential to assist users in interacting with complex software (e.g., PowerPoint, Photoshop). While prior research has primarily focused on automating user actions through clicks and keystrokes, this paradigm overlooks human intention, where users value the ability to explore, iterate, and refine their ideas while maintaining agency. To move beyond automation and toward collaboration, GUI agents must understand what users are doing and why. We introduce GUIDE (GUI User Intent Detection Evaluation), a benchmark that evaluates AI models on their ability to perceive user behavior, infer intent, and provide assistance in open-ended GUI tasks. GUIDE consists of 67.5 hours of screen recordings from 120 novice user demonstrations with think-aloud narrations, across 10 software. GUIDE defines three tasks - (i) Behavior State Detection, (ii) Intent Prediction, and (iii) Help Prediction that test a model's ability to recognize behavior state, reason about goals, and decide when and how to help. Evaluations across eight state-of-the-art multimodal models reveal that all models struggled, achieving only 44.6% and 55.0% accuracy on behavior state and help prediction. However, providing user context significantly improved the performance, raising help prediction by up to 50.2pp, highlighting the critical role of structured user understanding in effective assistance. Our dataset is available at https://guide-bench.github.io.

Haechan Kim, Junho Myung, Seoyoung Kim et al.

Prevalent ungrammatical expressions and disfluencies in spontaneous speech from second language (L2) learners pose unique challenges to Automatic Speech Recognition (ASR) systems. However, few datasets are tailored to L2 learner speech. We publicly release LearnerVoice, a dataset consisting of 50.04 hours of audio and transcriptions of L2 learners' spontaneous speech. Our linguistic analysis reveals that transcriptions in our dataset contain L2S (L2 learner's Spontaneous speech) features, consisting of ungrammatical expressions and disfluencies (e.g., filler words, word repetitions, self-repairs, false starts), significantly more than native speech datasets. Fine-tuning whisper-small.en with LearnerVoice achieves a WER of 10.26%, 44.2% lower than vanilla whisper-small.en. Furthermore, our qualitative analysis indicates that 54.2% of errors from the vanilla model on LearnerVoice are attributable to L2S features, with 48.1% of them being reduced in the fine-tuned model.

Tae Soo Kim, Yoonjoo Lee, Jaesang Yu et al.

To handle ambiguous and open-ended requests, Large Language Models (LLMs) are increasingly trained to interact with users to surface intents they have not yet expressed (e.g., ask clarification questions). However, users are often ambiguous because they have not yet formed their intents: they must observe and explore outcomes to discover what they want. Simply asking "what kind of tone do you want?" fails when users themselves do not know. We introduce DiscoverLLM, a novel and generalizable framework that trains LLMs to help users form and discover their intents. Central to our approach is a novel user simulator that models cognitive state with a hierarchy of intents that progressively concretize as the model surfaces relevant options -- where the degree of concretization serves as a reward signal that models can be trained to optimize. Resulting models learn to collaborate with users by adaptively diverging (i.e., explore options) when intents are unclear, and converging (i.e., refine and implement) when intents concretize. Across proposed interactive benchmarks in creative writing, technical writing, and SVG drawing, DiscoverLLM achieves over 10% higher task performance while reducing conversation length by up to 40%. In a user study with 75 human participants, DiscoverLLM improved conversation satisfaction and efficiency compared to baselines.

Kihoon Son, Hyewon Lee, DaEun Choi et al.

Human collaborators coordinate dynamically through process visibility and workspace awareness, yet AI agents typically either provide only final outputs or expose read-only execution processes (e.g., planning, reasoning) without interpreting concurrent user actions on shared artifacts. Building on mixed-initiative interaction principles, we explore whether agents can achieve collaborative context awareness -- interpreting concurrent user actions on shared artifacts and adapting in real-time. Study 1 (N=10 professional designers) revealed that process visibility enabled reasoning about agent actions but exposed conflicts when agents could not distinguish feedback from independent work. We developed CLEO, which interprets collaborative intent and adapts in real-time. Study 2 (N=10, two-day with stimulated recall interviews) analyzed 214 turns, identifying five action patterns, six triggers, and four enabling factors explaining when designers choose delegation (70.1%), direction (28.5%), or concurrent work (31.8%). We present a decision model with six interaction loops, design implications, and an annotated dataset.

Yoonsu Kim, Chanbin Park, Kihoon Son et al.

During complex knowledge work, people engage in iterative sensemaking: interpreting information, connecting ideas, and refining their understanding. Yet in current human-AI collaboration, these cognitive processes are difficult to share and organize for AI. They arise in situ and are rarely captured without interrupting the task, and even when expressed, remain scattered or reduced to system-generated summaries that fail to reflect users' cognitive processes. We address this challenge by enabling AI context that is grounded in users' cognitive traces and can be directly inspected and revised by the user. We first explore this through a probe system that supports in-situ snippet memoing, allowing users to easily share their cognitive moves. Our study (N=10) highlights the value of capturing such context and the challenge of organizing it once accumulated. We then present Contexty, which supports users in inspecting and refining these contexts to better reflect their understanding of the task. Our evaluation (N=12) showed that Contexty improved task awareness, thought structuring, and users' sense of authorship and control, with participants preferring snippet-grounded AI responses over non-grounded ones (78.1%). We discuss how capturing and organizing users' cognitive context enables AI as a context-aware collaborator while preserving user agency.

Juho Kim

In convolutional neural networks (CNNs), padding plays a pivotal role in preserving spatial dimensions throughout the layers. Traditional padding techniques do not explicitly distinguish between the actual image content and the padded regions, potentially causing CNNs to incorrectly interpret the boundary pixels or regions that resemble boundaries. This ambiguity can lead to suboptimal feature extraction. To address this, we propose PadChannel, a novel padding method that encodes padding statuses as an additional input channel, enabling CNNs to easily distinguish genuine pixels from padded ones. By incorporating PadChannel into several prominent CNN architectures, we observed small performance improvements and notable reductions in the variances on the ImageNet-1K image classification task at marginal increases in the computational cost. The source code is available at https://github.com/AussieSeaweed/pad-channel

Juho Kim, Tuomas Sandholm

Parallelization has played an instrumental role in the field of artificial intelligence (AI), drastically reducing the time taken to train and evaluate large AI models. In contrast to its impact in the broader field of AI, applying parallelization to computational game solving is relatively unexplored, despite its great potential. In this paper, we parallelize the family of counterfactual regret minimization (CFR) algorithms, which were central to important breakthroughs for solving large imperfect-information games. We present a generalized parallelization framework, reframing CFR as a series of linear algebra operations. Then, existing techniques for parallelizing linear algebra operations can be applied to accelerate CFR. We also describe how our technique can be applied to other tabular members of the CFR family of algorithms, including the state-of-the-art, such as CFR+, discounted CFR, and predictive variants of CFR. Experimentally, we show that our CFR implementation on a GPU is up to four orders of magnitude faster than Google DeepMind OpenSpiel's CFR implementations on a CPU.

Juho Kim, Tuomas Sandholm

Many games of interest in the real world are often intractably large, thereby necessitating the use of game abstraction to shrink them in size, typically by many magnitudes. Over the last two decades, there have been significant advances in game abstraction; however, the domain-specific nature (usually poker) of much of the prior work prevents those techniques from being easily generalized to other settings without extensively analyzing the game at hand. In this paper, we propose a domain-independent approach to game abstraction, which applies word embedding techniques from the field of natural language processing. Treating each action as a word and gameplay data as a corpus, word vectors can be trained to represent each action as a real-valued vector, which can then be clustered to facilitate game abstraction. We also explore the use of foundational embedding models and show that action embeddings obtained this way can capture a surprising amount of information about the underlying game. Experimental results demonstrate that our proposed game abstraction technique is effective, although it does not outperform specialized algorithms tailored to specific games.

Juho Kim, Fei Fang, Tuomas Sandholm

Watermarking techniques for large language models (LLMs), which encode hidden information in the output so its source can be verified, have gained significant attention in recent days, thanks to their potential capability to detect accidental or deliberate misuse. Similar challenges involving model misuse also exist in the context of game-playing, such as when detecting the unauthorized use of AI tools in gaming platforms (e.g., cheating in online chess). In this paper, we initiate the study of how game-playing strategies can be watermarked. We show how the KGW watermark for LLMs can be adapted to watermark game-playing agents in perfect-information extensive-form games. The watermark can then be detected using a statistical test. We show that the degradation in the quality of the watermarked strategy profile, quantified by the expected utility, can be bounded, but there is a tradeoff between detectability and quality. In our experiments, we bootstrap the watermarking framework to various chess engines and demonstrate that a) the impact of the watermark on the quality of the strategy is negligible and b) the watermark can be detected with just a handful of games.

Juho Kim, Tuomas Sandholm

How should an agent's performance in a multiagent environment be evaluated when there is a limited sample size or a high cost of running a trial? The AIVAT family of variance reduction techniques was proposed to address this challenge by introducing unbiased low-variance estimators of agents' expected payoffs. An important component of AIVAT is a heuristic value function that discriminates between potentially low- and high-value counterfactual histories. A notable gap in the literature is that there is little to no constraint or guideline on how the heuristic value function should be chosen or how uncertainty in its output should be handled. In our first contribution, we parameterize the heuristic value function to highlight AIVAT's potential vulnerabilities: a) the sample variance can be set pathologically low by directly applying gradient descent on the sample variance, and b) one can p-hack to draw a desired statistical conclusion via gradient descent/ascent on the test statistic. The main takeaway is that the heuristic value function should be fixed prior to observing the evaluation data! In our second contribution, we show how the heuristic uncertainty can be propagated to quantify the uncertainty of AIVAT estimates. It is then possible to further reduce the variance using inverse-variance weighted averaging, but AIVAT's unbiasedness guarantee may have to be sacrificed. In our experiments, we use a dataset of 10,000 poker hands to demonstrate our heuristic pathology and uncertainty results, with the latter yielding a 43.0% reduction in the number of samples (poker hands) needed to draw statistical conclusions.

Hyunseung Lim, Sooyohn Nam, Sungmin Na et al.

Patent examination remains an ongoing challenge in the NLP literature even after the advent of large language models (LLMs), as it requires an extensive yet nuanced human judgment on whether a submitted claim meets the statutory standards of novelty and non-obviousness against previously granted claims -- prior art -- in expert domains. Previous NLP studies have approached this challenge as a prediction task (e.g., forecasting grant outcomes) with high-level proxies such as similarity metrics or classifiers trained on historical labels. However, this approach often overlooks the step-by-step evaluations that examiners must make with profound information, including rationales for the decisions provided in office actions documents, which also makes it harder to measure the current state of techniques in patent review processes. To fill this gap, we construct PANORAMA, a dataset of 8,143 U.S. patent examination records that preserves the full decision trails, including original applications, all cited references, Non-Final Rejections, and Notices of Allowance. Also, PANORAMA decomposes the trails into sequential benchmarks that emulate patent professionals' patent review processes and allow researchers to examine large language models' capabilities at each step of them. Our findings indicate that, although LLMs are relatively effective at retrieving relevant prior art and pinpointing the pertinent paragraphs, they struggle to assess the novelty and non-obviousness of patent claims. We discuss these results and argue that advancing NLP, including LLMs, in the patent domain requires a deeper understanding of real-world patent examination. Our dataset is openly available at https://huggingface.co/datasets/LG-AI-Research/PANORAMA.

Ali Raeisdanaei, Juho Kim, Michael Liao et al.

In many safety-critical engineering domains, hazard analysis techniques are an essential part of requirement elicitation. Of the methods proposed for this task, STPA (System-Theoretic Process Analysis) represents a relatively recent development in the field. The completion, management, and traceability of this hazard analysis technique present a time-consuming challenge to the requirements and safety engineers involved. In this paper, we introduce a free, open-source software framework to build STPA models with several automated workflows powered by large language models (LLMs). In past works, LLMs have been successfully integrated into a myriad of workflows across various fields. Here, we demonstrate that LLMs can be used to complete tasks associated with STPA with a high degree of accuracy, saving the time and effort of the human engineers involved. We experimentally validate our method on real-world STPA models built by requirement engineers and researchers. The source code of our software framework is available at the following link: https://github.com/blueskysolarracing/stpa.

Juho Kim

This paper introduces the Poker Hand History (PHH) file format, designed to standardize the recording of poker hands across different game variants. Despite poker's widespread popularity in the mainstream culture as a mind sport and its prominence in the field of artificial intelligence (AI) research as a benchmark for imperfect information AI agents, it lacks a consistent format that humans can use to document poker hands across different variants that can also easily be parsed by machines. To address this gap in the literature, we propose the PHH format which provides a concise human-readable machine-friendly representation of hand history that comprehensively captures various details of the hand, ranging from initial game parameters and actions to contextual parameters including but not limited to the venue, players, and time control information. In the supplementary, we provide 10,088 hands covering 11 different variants in the PHH format. The full specification is available on https://github.com/uoftcprg/phh-std

Anjali Singh, Christopher Brooks, Warren Li et al.

Generating hints for incorrect code is a cognitively demanding task that fosters learning and metacognitive development. This study investigates three designs for personalized, scalable, and reflective hint-writing activities within a data science course: (i) writing a hint independently, (ii) writing a hint with on-demand AI assistance, and (iii) deferred AI assistance, in which students first write a hint independently and then revise it with the help of an AI-generated one. We examine how AI support can scaffold the learning process without diminishing students' productive cognitive effort. Through a randomized controlled experiment with graduate-level students (N=97), we found that deferring AI assistance leads to the highest-quality hints. Further, this design helps students identify a wide range of mistakes they otherwise struggle to identify without any AI assistance. Students valued these activities as opportunities to practice debugging and critically engage with AI outputs--skills that are now critical for learners to acquire as programming becomes increasingly automated and the use of AI for learning grows. Our findings also highlight key considerations for designing student-AI collaborative learning experiences to sustain student engagement, maintain appropriate cognitive load, and mitigate negative effects of AI, such as introducing redundancies and extraneous information into student work.

Yoonjoo Lee, Hyeonsu B. Kang, Matt Latzke et al. · allen-ai, cmu

With the rapid growth of scholarly archives, researchers subscribe to "paper alert" systems that periodically provide them with recommendations of recently published papers that are similar to previously collected papers. However, researchers sometimes struggle to make sense of nuanced connections between recommended papers and their own research context, as existing systems only present paper titles and abstracts. To help researchers spot these connections, we present PaperWeaver, an enriched paper alerts system that provides contextualized text descriptions of recommended papers based on user-collected papers. PaperWeaver employs a computational method based on Large Language Models (LLMs) to infer users' research interests from their collected papers, extract context-specific aspects of papers, and compare recommended and collected papers on these aspects. Our user study (N=15) showed that participants using PaperWeaver were able to better understand the relevance of recommended papers and triage them more confidently when compared to a baseline that presented the related work sections from recommended papers.

Bekzat Tilekbay, Saelyne Yang, Michal Lewkowicz et al.

Informational videos serve as a crucial source for explaining conceptual and procedural knowledge to novices and experts alike. When producing informational videos, editors edit videos by overlaying text/images or trimming footage to enhance the video quality and make it more engaging. However, video editing can be difficult and time-consuming, especially for novice video editors who often struggle with expressing and implementing their editing ideas. To address this challenge, we first explored how multimodality$-$natural language (NL) and sketching, which are natural modalities humans use for expression$-$can be utilized to support video editors in expressing video editing ideas. We gathered 176 multimodal expressions of editing commands from 10 video editors, which revealed the patterns of use of NL and sketching in describing edit intents. Based on the findings, we present ExpressEdit, a system that enables editing videos via NL text and sketching on the video frame. Powered by LLM and vision models, the system interprets (1) temporal, (2) spatial, and (3) operational references in an NL command and spatial references from sketching. The system implements the interpreted edits, which then the user can iterate on. An observational study (N=10) showed that ExpressEdit enhanced the ability of novice video editors to express and implement their edit ideas. The system allowed participants to perform edits more efficiently and generate more ideas by generating edits based on user's multimodal edit commands and supporting iterations on the editing commands. This work offers insights into the design of future multimodal interfaces and AI-based pipelines for video editing.

Hyungyu Shin, Jingyu Tang, Yoonjoo Lee et al.

Peer review underpins scientific progress, but it is increasingly strained by reviewer shortages and growing workloads. Large Language Models (LLMs) can automatically draft reviews now, but determining whether LLM-generated reviews are trustworthy requires systematic evaluation. Researchers have evaluated LLM reviews at either surface-level (e.g., BLEU and ROUGE) or content-level (e.g., specificity and factual accuracy). Yet it remains uncertain whether LLM-generated reviews attend to the same critical facets that human experts weigh -- the strengths and weaknesses that ultimately drive an accept-or-reject decision. We introduce a focus-level evaluation framework that operationalizes the focus as a normalized distribution of attention across predefined facets in paper reviews. Based on the framework, we developed an automatic focus-level evaluation pipeline based on two sets of facets: target (e.g., problem, method, and experiment) and aspect (e.g., validity, clarity, and novelty), leveraging 676 paper reviews (https://figshare.com/s/d5adf26c802527dd0f62) from OpenReview that consists of 3,657 strengths and weaknesses identified from human experts. The comparison of focus distributions between LLMs and human experts showed that the off-the-shelf LLMs consistently have a more biased focus towards examining technical validity while significantly overlooking novelty assessment when criticizing papers.

Minsun Kim, SeonGyeom Kim, Suyoun Lee et al.

This paper presents the development of a dashboard designed specifically for teachers in English as a Foreign Language (EFL) writing education. Leveraging LLMs, the dashboard facilitates the analysis of student interactions with an essay writing system, which integrates ChatGPT for real-time feedback. The dashboard aids teachers in monitoring student behavior, identifying noneducational interaction with ChatGPT, and aligning instructional strategies with learning objectives. By combining insights from NLP and Human-Computer Interaction (HCI), this study demonstrates how a human-centered approach can enhance the effectiveness of teacher dashboards, particularly in ChatGPT-integrated learning.

Benjamin Newman, Yoonjoo Lee, Aakanksha Naik et al. · allen-ai, uw

When conducting literature reviews, scientists often create literature review tables - tables whose rows are publications and whose columns constitute a schema, a set of aspects used to compare and contrast the papers. Can we automatically generate these tables using language models (LMs)? In this work, we introduce a framework that leverages LMs to perform this task by decomposing it into separate schema and value generation steps. To enable experimentation, we address two main challenges: First, we overcome a lack of high-quality datasets to benchmark table generation by curating and releasing arxivDIGESTables, a new dataset of 2,228 literature review tables extracted from ArXiv papers that synthesize a total of 7,542 research papers. Second, to support scalable evaluation of model generations against human-authored reference tables, we develop DecontextEval, an automatic evaluation method that aligns elements of tables with the same underlying aspects despite differing surface forms. Given these tools, we evaluate LMs' abilities to reconstruct reference tables, finding this task benefits from additional context to ground the generation (e.g. table captions, in-text references). Finally, through a human evaluation study we find that even when LMs fail to fully reconstruct a reference table, their generated novel aspects can still be useful.

Juho Kim

Counterfactual regret minimization is a family of algorithms of no-regret learning dynamics capable of solving large-scale imperfect information games. We propose implementing this algorithm as a series of dense and sparse matrix and vector operations, thereby making it highly parallelizable for a graphical processing unit, at a cost of higher memory usage. Our experiments show that our implementation performs up to about 401.2 times faster than OpenSpiel's Python implementation and, on an expanded set of games, up to about 203.6 times faster than OpenSpiel's C++ implementation and the speedup becomes more pronounced as the size of the game being solved grows.

Saelyne Yang, Anh Truong, Juho Kim et al.

Tutorial videos are a valuable resource for people looking to learn new tasks. People often learn these skills by viewing multiple tutorial videos to get an overall understanding of a task by looking at different approaches to achieve the task. However, navigating through multiple videos can be time-consuming and mentally demanding as these videos are scattered and not easy to skim. We propose VideoMix, a system that helps users gain a holistic understanding of a how-to task by aggregating information from multiple videos on the task. Insights from our formative study (N=12) reveal that learners value understanding potential outcomes, required materials, alternative methods, and important details shared by different videos. Powered by a Vision-Language Model pipeline, VideoMix extracts and organizes this information, presenting concise textual summaries alongside relevant video clips, enabling users to quickly digest and navigate the content. A comparative user study (N=12) demonstrated that VideoMix enabled participants to gain a more comprehensive understanding of tasks with greater efficiency than a baseline video interface, where videos are viewed independently. Our findings highlight the potential of a task-oriented, multi-video approach where videos are organized around a shared goal, offering an enhanced alternative to conventional video-based learning.

Hyoungwook Jin, Yoonsu Kim, Dongyun Jung et al.

Mathematics learning entails mastery of both content knowledge and cognitive processing of knowing, applying, and reasoning with it. Automated math assessment primarily has focused on grading students' exhibition of content knowledge by finding textual evidence, such as specific numbers, formulas, and statements. Recent advancements in problem-solving, image recognition, and reasoning capabilities of large language models (LLMs) show promise for nuanced evaluation of students' cognitive skills. Diagnosing cognitive skills needs to infer students' thinking processes beyond textual evidence, which is an underexplored task in LLM-based automated assessment. In this work, we investigate how state-of-the-art LLMs diagnose students' cognitive skills in mathematics. We constructed MathCog, a novel benchmark dataset comprising 639 student responses to 110 expert-curated middle school math problems, each annotated with detailed teachers' diagnoses based on cognitive skill checklists. Using MathCog, we evaluated 16 closed and open LLMs of varying model sizes and vendors. Our evaluation reveals that even the state-of-the-art LLMs struggle with the task, all F1 scores below 0.5, and tend to exhibit strong false confidence for incorrect cases ($r_s=.617$). We also found that model size positively correlates with the diagnosis performance ($r_s=.771$). Finally, we discuss the implications of these findings, the overconfidence issue, and directions for improving automated cognitive skill diagnosis.

Kihoon Son, DaEun Choi, Tae Soo Kim et al.

Capturing professionals' decision-making in creative workflows is essential for reflection, collaboration, and knowledge sharing, yet existing methods often leave rationales incomplete and implicit decisions hidden. To address this, we present CLEAR framework that structures reasoning into cognitive decision steps-linked units of actions, artifacts, and self-explanations that make decisions traceable. Building on this framework, we introduce ClearFairy, a think-aloud AI assistant for UI design that detects weak explanations, asks lightweight clarifying questions, and infers missing rationales to ease the knowledge-sharing burden. In a study with twelve creative professionals, 85% of ClearFairy's inferred rationales were accepted, increasing strong explanations from 14% to over 83% of decision steps without adding cognitive demand. The captured steps also enhanced generative AI agents in Figma, yielding next-action predictions better aligned with professionals and producing more coherent design outcomes. For future research on human knowledge-grounded creative AI agents, we release a dataset of captured 417 decision steps.

Juho Kim, Michael Guerzhoy

A \textit{culture of honor} refers to a social system where individuals' status, reputation, and esteem play a central role in governing interpersonal relations. Past works have associated this concept with the United States (US) South and related with it various traits such as higher sensitivity to insult, a higher value on reputation, and a tendency to react violently to insults. In this paper, we hypothesize and confirm that internet users from the US South, where a \textit{culture of honor} is more prevalent, are more likely to display a trait predicted by their belonging to a \textit{culture of honor}. Specifically, we test the hypothesis that US Southerners are more likely to retaliate to personal attacks by personally attacking back. We leverage OpenAI's GPT-3.5 API to both geolocate internet users and to automatically detect whether users are insulting each other. We validate the use of GPT-3.5 by measuring its performance on manually-labeled subsets of the data. Our work demonstrates the potential of formulating a hypothesis based on a conceptual framework, operationalizing it in a way that is amenable to large-scale LLM-aided analysis, manually validating the use of the LLM, and drawing a conclusion.

Daeheon Jeong, Seoyeon Byun, Kihoon Son et al.

User interface (UI) design is an iterative process in which designers progressively refine their work with design software such as Figma or Sketch. Recent advances in vision language models (VLMs) with tool invocation suggest these models can operate design software to edit a UI design through iteration. Understanding and enhancing this capacity is important, as it highlights VLMs' potential to collaborate with designers within conventional software. However, as no existing benchmark evaluates tool-based design performance, the capacity remains unknown. To address this, we introduce CANVAS, a benchmark for VLMs on tool-based user interface design. Our benchmark contains 598 tool-based design tasks paired with ground-truth references sampled from 3.3K mobile UI designs across 30 function-based categories (e.g., onboarding, messaging). In each task, a VLM updates the design step-by-step through context-based tool invocations (e.g., create a rectangle as a button background), linked to design software. Specifically, CANVAS incorporates two task types: (i) design replication evaluates the ability to reproduce a whole UI screen; (ii) design modification evaluates the ability to modify a specific part of an existing screen. Results suggest that leading models exhibit more strategic tool invocations, improving design quality. Furthermore, we identify common error patterns models exhibit, guiding future work in enhancing tool-based design capabilities.

Tae Soo Kim, Heechan Lee, Yoonjoo Lee et al.

Practitioners increasingly rely on Large Language Models (LLMs) to evaluate generative AI outputs through "LLM-as-a-Judge" approaches. However, these methods produce holistic scores that obscure which specific elements influenced the assessments. We propose functional fragmentation, a method that dissects each output into key fragments and interprets the rhetoric functions that each fragment serves relative to evaluation criteria -- surfacing the elements of interest and revealing how they fulfill or hinder user goals. We instantiate this approach in Evalet, an interactive system that visualizes fragment-level functions across many outputs to support inspection, rating, and comparison of evaluations. A user study (N=10) found that, while practitioners struggled to validate holistic scores, our approach helped them identify 48% more evaluation misalignments. This helped them calibrate trust in LLM evaluations and rely on them to find more actionable issues in model outputs. Our work shifts LLM evaluation from quantitative scores toward qualitative, fine-grained analysis of model behavior.

Tae Soo Kim, Yoonjoo Lee, Yoonah Park et al.

Personalization of Large Language Models (LLMs) often assumes users hold static preferences that reflect globally in all tasks. In reality, humans hold dynamic preferences that change depending on the context. As users interact with an LLM in various contexts, they naturally reveal their contextual preferences, which a model must infer and apply in future contexts to ensure alignment. To assess this, we introduce CUPID, a benchmark of 756 human-curated interaction session histories between users and LLM-based chat assistants. In each interaction session, the user provides a request in a specific context and expresses their preference through multi-turn feedback. Given a new user request and prior interaction sessions, our benchmark assesses whether LLMs can infer the preference relevant to this request and generate a response that satisfies this preference. With CUPID, we evaluated 10 open and proprietary LLMs, revealing that state-of-the-art LLMs struggle to infer preferences from multi-turn interactions and fail to discern what previous context is relevant to a new request -- under 50% precision and 65% recall. Our work highlights the need to advance LLM capabilities for more contextually personalized interactions and proposes CUPID as a resource to drive these improvements.

Youngmin Seo, Yunhyeong Kwon, Younghun Park et al.

Accurate spatial prediction of power integrity issues, such as IR drop, is critical for reliable VLSI design. However, traditional simulation-based solvers are computationally expensive and difficult to scale. We address this challenge by reformulating IR drop estimation as a pixel-wise regression task on heterogeneous multi-channel physical maps derived from circuit layouts. Prior learning-based methods treat all input layers (e.g., metal, via, and current maps) equally, ignoring their varying importance to prediction accuracy. To tackle this, we propose a novel Weakness-Aware Channel Attention (WACA) mechanism, which recursively enhances weak feature channels while suppressing over-dominant ones through a two-stage gating strategy. Integrated into a ConvNeXtV2-based attention U-Net, our approach enables adaptive and balanced feature representation. On the public ICCAD-2023 benchmark, our method outperforms the ICCAD-2023 contest winner by reducing mean absolute error by 61.1% and improving F1-score by 71.0%. These results demonstrate that channel-wise heterogeneity is a key inductive bias in physical layout analysis for VLSI.

Juho Kim

The independent chip model (ICM) forms a cornerstone of all modern poker tournament strategy. However, despite its prominence, the ICM's performance in the real world has not been sufficiently scrutinized, especially at a large scale. In this paper, we introduce our new dataset of poker tournaments, consisting of results of over ten thousand events. Then, using this dataset, we perform two experiments as part of a large-scale empirical validation of the ICM. First, we verify that the ICM performs more accurately than a baseline we propose. Second, we obtain empirical evidence of the ICM underestimating the performances of players with larger stacks while overestimating those who are short-stacked. Our contributions may be useful to future researchers developing new algorithms for estimating a player's value in poker tournaments.

Brandon Lepine, Gawesha Weerantunga, Juho Kim et al.

Current AI benchmarks miss the messy, multi-turn nature of human-AI collaboration. We present an evaluation framework that decomposes real-world tasks into interdependent subtasks, letting us track both LLM performance and users' strategies across a dialogue. Complementing this framework, we develop a suite of metrics, including a composite usage derived from semantic similarity, word overlap, and numerical matches; structural coherence; intra-turn diversity; and a novel measure of the "information frontier" reflecting the alignment between AI outputs and users' working knowledge. We demonstrate our methodology in a financial valuation task that mirrors real-world complexity. Our empirical findings reveal that while greater integration of LLM-generated content generally enhances output quality, its benefits are moderated by factors such as response incoherence, excessive subtask diversity, and the distance of provided information from users' existing knowledge. These results suggest that proactive dialogue strategies designed to inject novelty may inadvertently undermine task performance. Our work thus advances a more holistic evaluation of human-AI collaboration, offering both a robust methodological framework and actionable insights for developing more effective AI-augmented work processes.

Gionnieve Lim, Juho Kim, Simon T. Perrault

Social platforms have expanded opportunities for deliberation with the comments being used to inform one's opinion. However, using such information to form opinions is challenged by unsubstantiated or false content. To enhance the quality of opinion formation and potentially confer resistance to misinformation, we developed Iffy-Or-Not (ION), a browser extension that seeks to invoke critical thinking when reading texts. With three features guided by argumentation theory, ION highlights fallacious content, suggests diverse queries to probe them with, and offers deeper questions to consider and chat with others about. From a user study (N=18), we found that ION encourages users to be more attentive to the content, suggests queries that align with or are preferable to their own, and poses thought-provoking questions that expands their perspectives. However, some participants expressed aversion to ION due to misalignments with their information goals and thinking predispositions. Potential backfiring effects with ION are discussed.

Yoonsu Kim, Kihoon Son, Seoyoung Kim et al.

AI intent alignment, ensuring that AI produces outcomes as intended by users, is a critical challenge in human-AI interaction. The emergence of generative AI, including LLMs, has intensified the significance of this problem, as interactions increasingly involve users specifying desired results for AI systems. In order to support better AI intent alignment, we aim to explore human strategies for intent specification in human-human communication. By studying and comparing human-human and human-LLM communication, we identify key strategies that can be applied to the design of AI systems that are more effective at understanding and aligning with user intent. This study aims to advance toward a human-centered AI system by bringing together human communication strategies for the design of AI systems.

Yoonjoo Lee, Kihoon Son, Tae Soo Kim et al.

As Large Language Models (LLMs) are nondeterministic, the same input can generate different outputs, some of which may be incorrect or hallucinated. If run again, the LLM may correct itself and produce the correct answer. Unfortunately, most LLM-powered systems resort to single results which, correct or not, users accept. Having the LLM produce multiple outputs may help identify disagreements or alternatives. However, it is not obvious how the user will interpret conflicts or inconsistencies. To this end, we investigate how users perceive the AI model and comprehend the generated information when they receive multiple, potentially inconsistent, outputs. Through a preliminary study, we identified five types of output inconsistencies. Based on these categories, we conducted a study (N=252) in which participants were given one or more LLM-generated passages to an information-seeking question. We found that inconsistency within multiple LLM-generated outputs lowered the participants' perceived AI capacity, while also increasing their comprehension of the given information. Specifically, we observed that this positive effect of inconsistencies was most significant for participants who read two passages, compared to those who read three. Based on these findings, we present design implications that, instead of regarding LLM output inconsistencies as a drawback, we can reveal the potential inconsistencies to transparently indicate the limitations of these models and promote critical LLM usage.

Meng Xia, Yankun Zhao, Mehmet Hamza Erol et al.

With the rise of the gig economy, online language tutoring platforms are becoming increasingly popular. They provide temporary and flexible jobs for native speakers as tutors and allow language learners to have one-on-one speaking practices on demand. However, the lack of stable relationships hinders tutors and learners from building long-term trust. "Distributed tutorship" -- temporally discontinuous learning experience with different tutors -- has been underexplored yet has many implications for modern learning platforms. In this paper, we analyzed tutorship sequences of 15,959 learners and found that around 40% of learners change to new tutors every session; 44% learners change to new tutors while reverting to previous tutors sometimes; only 16% learners change to new tutors and then fix on one tutor. We also found suggestive evidence that higher distributedness -- higher diversity and lower continuity in tutorship -- is correlated to slower improvements in speaking performance scores with a similar number of sessions. We further surveyed 519 and interviewed 40 learners and found that more learners preferred fixed tutorship while some do not have it due to various reasons. Finally, we conducted semi-structured interviews with three tutors and one product manager to discuss the implications for improving the continuity in learning under distributed tutorship.

Juho Kim, Christopher Malon, Asim Kadav

Existing entailment datasets mainly pose problems which can be answered without attention to grammar or word order. Learning syntax requires comparing examples where different grammar and word order change the desired classification. We introduce several datasets based on synthetic transformations of natural entailment examples in SNLI or FEVER, to teach aspects of grammar and word order. We show that without retraining, popular entailment models are unaware that these syntactic differences change meaning. With retraining, some but not all popular entailment models can learn to compare the syntax properly.

Keunwoo Choi, Deokjin Joo, Juho Kim

We introduce Kapre, Keras layers for audio and music signal preprocessing. Music research using deep neural networks requires a heavy and tedious preprocessing stage, for which audio processing parameters are often ignored in parameter optimisation. To solve this problem, Kapre implements time-frequency conversions, normalisation, and data augmentation as Keras layers. We report simple benchmark results, showing real-time on-GPU preprocessing adds a reasonable amount of computation.

Juho Kim, Elena L. Glassman, Andrés Monroy-Hernández et al.

Teachers in conventional classrooms often ask learners to express themselves and show their thought processes by speaking out loud, drawing on a whiteboard, or even using physical objects. Despite the pedagogical value of such activities, interactive exercises available in most online learning platforms are constrained to multiple-choice and short answer questions. We introduce RIMES, a system for easily authoring, recording, and reviewing interactive multimedia exercises embedded in lecture videos. With RIMES, teachers can prompt learners to record their responses to an activity using video, audio, and inking while watching lecture videos. Teachers can then review and interact with all the learners' responses in an aggregated gallery. We evaluated RIMES with 19 teachers and 25 students. Teachers created a diverse set of activities across multiple subjects that tested deep conceptual and procedural knowledge. Teachers found the exercises useful for capturing students' thought processes, identifying misconceptions, and engaging students with content.

Elena L. Glassman, Juho Kim, Andrés Monroy-Hernández et al.

Educators have developed an effective technique to get feedback after in-person lectures, called "muddy card." Students are given time to reflect and write the "muddiest" (least clear) point on an index card, to hand in as they leave class. This practice of assigning end-of-lecture reflection tasks to generate explicit student feedback is well suited for adaptation to the challenge of supporting feedback in online video lectures. We describe the design and evaluation of Mudslide, a prototype system that translates the practice of muddy cards into the realm of online lecture videos. Based on an in-lab study of students and teachers, we find that spatially contextualizing students' muddy point feedback with respect to particular lecture slides is advantageous to both students and teachers. We also reflect on further opportunities for enhancing this feedback method based on teachers' and students' experiences with our prototype.

Joseph Jay Williams, Juho Kim, Brian C. Keegan

Most education and workplace learning takes place in classroom contexts far removed from laboratories or field sites with special arrangements for scientific research. But digital online resources provide a novel opportunity for large scale efforts to bridge the real world and laboratory settings which support data collection and randomized A/B experiments comparing different versions of content or interactions [2]. However, there are substantial technological and practical barriers in aligning instructors and researchers to use learning technologies like blended lessons/exercises & MOOCs as both a service for students and a realistic context to conduct research. This paper explains how the concept of a MOOClet can facilitate research-practitioner collaborations. MOOClets [3] are defined as modular components of a digital resource that can be implemented in technology to: (1) allow modification to create multiple versions, (2) allow experimental comparison and personalization of different versions, (3) reliably specify what data are collected. We suggest a framework in which instructors specify what kinds of changes to lessons, exercises, and emails they would be willing to adopt, and what data they will collect and make available. Researchers can then: (1) specify or design experiments that compare the effects of different versions on quantifiable outcomes. (2) Explore algorithms for maximizing particular outcomes by choosing alternative versions of a MOOClet based on the input variables available. We present a prototype survey tool for instructors intended to facilitate practitioner researcher matches and successful collaborations.

Joseph Jay Williams, Korinn Ostrow, Xiaolu Xiong et al.

In contrast to typical laboratory experiments, the everyday use of online educational resources by large populations and the prevalence of software infrastructure for A/B testing leads us to consider how platforms can embed in vivo experiments that do not merely support research, but ensure practical improvements to their educational components. Examples are presented of randomized experimental comparisons conducted by subsets of the authors in three widely used online educational platforms Khan Academy, edX, and ASSISTments. We suggest design principles for platform technology to support randomized experiments that lead to practical improvements enabling Iterative Improvement and Collaborative Work and explain the benefit of their implementation by WPI co-authors in the ASSISTments platform.

Anant Bhardwaj, Juho Kim, Steven Dow et al.

Constructing a good conference schedule for a large multi-track conference needs to take into account the preferences and constraints of organizers, authors, and attendees. Creating a schedule which has fewer conflicts for authors and attendees, and thematically coherent sessions is a challenging task. Cobi introduced an alternative approach to conference scheduling by engaging the community to play an active role in the planning process. The current Cobi pipeline consists of committee-sourcing and author-sourcing to plan a conference schedule. We further explore the design space of community-sourcing by introducing attendee-sourcing -- a process that collects input from conference attendees and encodes them as preferences and constraints for creating sessions and schedule. For CHI 2014, a large multi-track conference in human-computer interaction with more than 3,000 attendees and 1,000 authors, we collected attendees' preferences by making available all the accepted papers at the conference on a paper recommendation tool we built called Confer, for a period of 45 days before announcing the conference program (sessions and schedule). We compare the preferences marked on Confer with the preferences collected from Cobi's author-sourcing approach. We show that attendee-sourcing can provide insights beyond what can be discovered by author-sourcing. For CHI 2014, the results show value in the method and attendees' participation. It produces data that provides more alternatives in scheduling and complements data collected from other methods for creating coherent sessions and reducing conflicts.