Pith. sign in

REVIEW 4 major objections 6 minor 69 references

DxHF: Providing High-Quality Human Feedback for LLM Alignment via Interactive Decomposition

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Decomposing LLM responses into atomic claims improves the accuracy of human preference feedback, especially for uncertain annotators.

desk verdict A genuinely useful interface study with a modest but real effect, undercut by a missing bias check on the LLM-generated claims. read the letter →

arxiv 2507.18802 v1 pith:X5GO5OFM submitted 2025-07-24 cs.HC cs.AI

Danqing Shi , Furui Cheng , Tino Weinkauf , Antti Oulasvirta , Mennatallah El-Assady This is my paper · ORCID
classification cs.HCcs.AI
keywords AIalignmenthumanfeedbackRLHFdecompositionatomicclaimsuserinterfacecrowdsourcingLLMevaluation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that the bottleneck in RLHF-style alignment is not the learning algorithm but the interface through which humans express preferences. Its central claim is that decomposing two long LLM responses into atomic claims—single-information sentences—and presenting them with relevance ranking and semantic links improves the accuracy of pairwise preference judgments. In a crowdsourced study of 160 participants, the DxHF interface reached 66.3% agreement with the reference labels versus 61.6% for a plain side-by-side baseline, a gain driven by annotators who reported low certainty. The cost is an average of 18 additional seconds per judgment. If the finding holds, better annotation interfaces are a direct lever on alignment quality.

What carries the argument

The load-bearing mechanism is decomposition of long responses into atomic claims, following the idea that dividing a judgment into smaller sub-judgments improves accuracy. Each response is broken into claims by GPT-4, copying words from the original text; a Cross-Encoder scores each claim's relevance to the conversation query, and claims are linked across the two responses using cosine similarity of BERT embeddings, with a keyword label summarizing each linked pair. The DxHF interface displays the two claim lists side by side with relevance encoded as text opacity, hover highlighting to reveal linked claims, and accordion folding so annotators can toggle between full text and decomposition.

What would settle it

A direct test would recruit an independent expert panel to label the same 50-100 comparison pairs, then measure whether DxHF's accuracy advantage over the baseline persists when the reference labels come from the experts rather than from HH-RLHF. A second test would replace GPT-4's decomposition with a randomized or naive sentence splitter; if the accuracy gain disappears, the effect is attributable to the decomposition fidelity itself rather than the interface layout.

Watch

Extended reading notes

Core claim

On the paper's own terms, DxHF establishes that the decomposition principle improves human comparative feedback for LLM alignment. The user study reports 66.3% accuracy with DxHF versus 61.6% with a baseline that mirrors the standard ChatGPT-style comparison page (Wilcoxon signed-rank, p=0.0293). For the 138 participants who rated their certainty below 5, accuracy is 6.4 percentage points higher with DxHF (p=0.0059), matching the simulation prediction that decomposition helps most when judges are uncertain. A separate ablation study with 36 participants finds the full interface is rated most useful and scores highest accuracy, with ranking supporting attention and linking reducing comparison effort. The paper also acknowledges the design is suited to factual or task-oriented comparisons, not holistic judgments of coherence, tone, or style.

Load-bearing premise

The evaluation treats the HH-RLHF crowd labels as ground truth for helpfulness, so the reported accuracy is agreement with those labels rather than an objective measure of better feedback, and it further assumes that GPT-4's decomposition preserves the original text's meaning.

Editorial extensions

If this is right

  • Preference data for RLHF and DPO can be made more accurate by changing only the annotation display, without modifying the learning algorithm.
  • Low-certainty annotators, who are common on crowdsourcing platforms, receive the largest accuracy boost from decomposition; the paper reports 6.4% higher accuracy for this group.
  • The average 18-second increase in feedback time is an explicit speed-accuracy trade-off that practitioners can adjust by task complexity.
  • The ablation study indicates both design features earn their place: ranking focuses attention on key claims, and linking lowers the effort needed to compare across the two responses.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the reported agreement with HH-RLHF labels reflects genuine judgment quality, reward models trained on DxHF-collected preferences should produce measurably different alignment outcomes than models trained on baseline preferences; this is a direct, testable consequence.
  • The method's ceiling is set by decomposition fidelity, so adding an automatic check that each claim is semantically entailed by its source sentence would separate the interface's contribution from the decomposer's.
  • The same decomposition-plus-linking display could be applied to other pairwise text judgments, such as comparing two news articles or judging whether an AI edit preserved a writer's intent, wherever the texts are long enough to overwhelm working memory.
  • For expert annotators who are already confident, the benefit may be small, meaning the interface is most valuable in high-throughput crowd pipelines with heterogeneous effort rather than in expert review.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. This paper proposes DxHF, a user interface for pairwise comparative feedback in LLM alignment. DxHF decomposes each long-form response into atomic claims via GPT-4, ranks claims by query relevance using a cross-encoder, links similar claims across responses with summary keywords, and visually encodes relevance with opacity and hover highlighting. The authors evaluate DxHF in three ways: a simulation with LLM-as-a-judge synthetic annotators over HH-RLHF data, a within-subjects crowdsourcing study with 160 Prolific participants on 10 tasks, and an ablation study with 36 participants comparing full DxHF with versions lacking linking or ranking. The user study reports 66.3% versus 61.6% accuracy relative to HH-RLHF labels (p=0.0293), with a larger 6.4% advantage for the 138 participants with lower certainty (p=0.0059), at a cost of 18 seconds of additional average feedback time. The central claim is that the decomposition principle improves the quality of human comparative feedback for LLM alignment.

Significance. If the central claim holds, the paper makes a useful HCI contribution to LLM alignment: it demonstrates that interface design can improve preference data quality, and the tool is open-sourced. The study has real strengths: a genuine crowdsourcing experiment with native speakers, counterbalanced interface order, a significant main effect and a significant subgroup effect for uncertain annotators, an ablation with a statistically tested usefulness rating, and an unusually candid discussion of the risk that the LLM-based pipeline introduces its own bias. However, the evidence as presented does not yet separate the effect of decomposition from machine-generated attention cueing, and the metric is agreement with crowd labels rather than an independent measure of feedback quality. The paper's value therefore depends on additional validity checks.

major comments (4)
  1. [Sec 6.1-6.2] The main accuracy result treats HH-RLHF preference labels as ground truth. These labels are crowd preferences, not objective ground truth, so "accuracy" in Fig. 6 is agreement with the original crowd. Because the DxHF pipeline (GPT-4 decomposition, cross-encoder relevance scoring, keyword linking; Sec 4.2) may systematically surface cues that correlate with those labels, the 4.7% gain could reflect participants following machine-selected highlights rather than providing better independent judgments. The paper itself concedes in Sec 8 that DxHF "might introduce its own bias... attributable to the LLM that is used to pick the highlighted points," and in Sec 4.2 it concedes the decomposition "may occasionally add or modify words," yet no fidelity or bias check is reported. Please provide a no-highlight/no-decomposition control, an independent expert-label evaluation on a subset of items, or a per-task analysis showing that the improvement is not driven by items where the pipeline's aggregate claim scores predict the HH-RLHF label.
  2. [Sec 7] The ablation cannot attribute the accuracy gain to decomposition. The two ablated conditions ("w/o linking" and "w/o ranking") keep the decomposed claims; there is no condition that presents the original text with the same highlighting and linking, and no no-highlight control. In addition, "w/o ranking" removes only the color encoding, not the relevance-sorting interaction, so it is not a clean ablation of ranking. Accuracy differences (66.7%, 65.7%, 63.%) are reported without inferential statistics; only the usefulness ANOVA is tested. Please report statistical tests for accuracy and include a control that removes decomposition while retaining the visual cues, or the conclusion that decomposition itself improves feedback quality is not supported.
  3. [Sec 5] The simulation uses GPT-4 both to decompose the text and to judge helpfulness of full texts and claims, and it measures agreement with HH-RLHF labels. The advantage of the decomposition strategies in Fig. 5 could therefore be an artifact of the judge scoring decomposed claims more consistently with the HH-RLHF labels than full texts, rather than evidence about human comparison behavior. The statement in Sec 5.3 that the simulation "theoretically confirms" the advantage is overstated for an LLM simulation; please validate the simulated strategies against the user-study data or present the simulation only as a hypothesis-generating check.
  4. [Sec 6.1] The paper does not report how the 10 tasks were assigned to the two interface conditions. It states that the order of interfaces was counterbalanced, but not that task assignment was counterbalanced or randomized. If the same five tasks were always paired with one interface, task difficulty differences alone could produce the observed 4.7% gap. Please clarify the task-condition assignment and report per-task accuracy or a mixed-effects model with task as a random effect.
minor comments (6)
  1. [Sec 2.1] The phrase "highly relys" should be "highly rely."
  2. [Sec 3.2] The sentence "The key challenge in achieving this goal is that is that comparing long text snippets" contains a duplicated "is that."
  3. [Sec 7.2] "63.%" appears to be missing a digit, and "Partcipant" should be "Participant."
  4. [Sec 5.2] The sampling procedure says 50 instances are sampled, but the user study uses 10; please report how many were excluded by the domain-knowledge filter and how the final 10 tasks were chosen.
  5. [Sec 6.1] The number of participants excluded for responding in under 10 seconds is not reported; please state it and note whether the results are sensitive to this exclusion.
  6. [Sec 2.1] The claim that there are no existing studies on improving human feedback quality "beyond simply comparing two texts" is too strong given the cited MOCHA and LLM Comparator work; please qualify the statement.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the main claim rests on a human user study with external ground truth, and no prediction reduces to its inputs by construction.

full rationale

The paper's central claim—that DxHF improves human comparative feedback accuracy—is established through a crowdsourcing study where 160 participants compare responses using DxHF versus a baseline, with accuracy measured as agreement with HH-RLHF human preference labels (Sec 6.2). This is an independent empirical evaluation, not a quantity derived from fitted parameters or from the model's own output. The technical evaluation in Sec 5 uses GPT-4 as a simulated judge for both full-text and decomposed-claim scoring, which is a modeling choice with possible bias, but it is not a prediction that reduces by construction: the comparison is between LLM-scores under different presentation strategies, and the main claims do not depend on this simulation alone. No equation in the paper defines the outcome in terms of the input. The decomposition, ranking, and linking modules (Sec 4.2) are described as design features; their contributions are tested in the ablation study and user study rather than assumed. Self-citations appear only in related work (e.g., Interactive-RLHF, Interactive Reward Tuning, RELIC) and are descriptive rather than load-bearing. The paper also explicitly acknowledges a potential bias from the LLM-based highlighting in Sec 8, which is a limitation statement rather than a circular derivation. No fitted input is renamed as a prediction, and no uniqueness theorem or prior same-author result is invoked to force the design. Therefore, the derivation chain is self-contained with respect to the central empirical claim, and no circular step is present.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central claim rests on several unverified domain assumptions: the validity of HH-RLHF labels as ground truth, the faithfulness of LLM-based decomposition, and the transfer of the decomposition principle to LLM text comparison. Numeric thresholds for ranking, linking, and data filtering are hand-chosen and can affect the reported effect sizes.

free parameters (4)
  • Relevance threshold t_h = 0.3
    Hand-chosen threshold in technical evaluation (Sec 5.1) to filter out less relevant claims; affects all decomposition+ranking results.
  • Linking similarity threshold = 0.7
    Hand-chosen threshold (Sec 5.1) to include only high-similarity linked claims in the simulation strategy.
  • Minimum sentence count for inclusion = 5 sentences
    Data selection filter (Sec 5.2) excluding short responses; affects the sample.
  • Maximum length discrepancy = 30 words
    Data selection filter (Sec 5.2) excluding pairs with large length differences.
assumptions (6)
  • domain assumption HH-RLHF dataset labels are treated as ground truth for helpfulness.
    Used as the accuracy target in the user study (Sec 5.2, Sec 6.1); if these labels are noisy or biased, the reported accuracy improvement is agreement with that crowd, not objective quality.
  • domain assumption LLM-based decomposition preserves the information content of the original text.
    Section 4.2 claims fidelity to source words, but notes the model may occasionally add or modify words; no human evaluation of decomposition faithfulness.
  • domain assumption LLM-as-a-judge scores correlate with human judgments of helpfulness.
    Technical evaluation relies on GPT-4 scoring of claims and texts (Sec 5.1, Appendix A.3), citing prior work for correlation.
  • domain assumption The decomposition principle (Armstrong 1975) transfers to LLM text comparison.
    The paper builds its design on this principle (Sec 4.1) and interprets results through it.
  • domain assumption Boltzmann rationality models annotator decision noise.
    Used in simulation (Sec 5.1, Eq. 1) to vary user capability.
  • domain assumption The 10 selected tasks are representative of LLM alignment comparison tasks.
    User study uses a small filtered subset of HH-RLHF (Sec 6.1).

how reviews work

0 comments
Cite this review

Pith. "Pith review of DxHF: Providing High-Quality Human Feedback for LLM Alignment via Interactive Decomposition." pith.science (2026). https://pith.science/paper/X5GO5OFM

@misc{pith2026250718802,
  author       = {Pith},
  title        = {Pith review of: DxHF: Providing High-Quality Human Feedback for LLM Alignment via Interactive Decomposition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/X5GO5OFM}},
  note         = {Machine review of arXiv:2507.18802}
}
read the original abstract

Human preferences are widely used to align large language models (LLMs) through methods such as reinforcement learning from human feedback (RLHF). However, the current user interfaces require annotators to compare text paragraphs, which is cognitively challenging when the texts are long or unfamiliar. This paper contributes by studying the decomposition principle as an approach to improving the quality of human feedback for LLM alignment. This approach breaks down the text into individual claims instead of directly comparing two long-form text responses. Based on the principle, we build a novel user interface DxHF. It enhances the comparison process by showing decomposed claims, visually encoding the relevance of claims to the conversation and linking similar claims. This allows users to skim through key information and identify differences for better and quicker judgment. Our technical evaluation shows evidence that decomposition generally improves feedback accuracy regarding the ground truth, particularly for users with uncertainty. A crowdsourcing study with 160 participants indicates that using DxHF improves feedback accuracy by an average of 5%, although it increases the average feedback time by 18 seconds. Notably, accuracy is significantly higher in situations where users have less certainty. The finding of the study highlights the potential of HCI as an effective method for improving human-AI alignment.

Figures

Figures reproduced from arXiv: 2507.18802 by the authors.

Figure 1
Figure 1. An illustration of the comparison between stan [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The screenshots of LLM alignment user interfaces. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. DxHF decomposes the text and organizes them into individual claims. Similar claims across two responses are connected with a keyword label. By using hover highlights, human annotators can more easily identify differences and compare the claims. We provide two examples: (a) grouping the links (≡) and hovering over a keyword, and (b) sorting the claims (↑↓) and hovering over a specific claim [PITH_FULL_IMAGE:figures/… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: The design is based on a visual metaphor of an [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Accuracy vs. rationality of simulated annotators, [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Accuracy and speed analysis of the crowdsourcing [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: The ablation study results illustrate that [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: The alternative designs for the claim annotations [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]
Figure 9
Figure 9. Figure 9: The screenshot of the user interface used in the [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

69 extracted references · 43 canonical work pages

  1. [1]

    Chat GPT now asking casual users to choose the better of two responses

    2024. Chat GPT now asking casual users to choose the better of two responses . https://www.reddit.com/r/DataAnnotationTech/comments/1ge0pwf

    work page 2024

  2. [2]

    Human preference data about helpfulness and harmlessness

    2025. Human preference data about helpfulness and harmlessness . https: //huggingface.co/datasets/Anthropic/hh-rlhf

    work page 2025

  3. [3]

    Synthetic Users: Run your user and market research with the most human-like AI participants

    2025. Synthetic Users: Run your user and market research with the most human-like AI participants. https://www.syntheticusers.com/

    work page 2025

  4. [4]

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Floren- cia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774 (2023)

    arXiv 2023

  5. [5]

    Ian Arawjo, Chelse Swoopes, Priyan Vaithilingam, Martin Wattenberg, and Elena L Glassman. 2024. ChainForge: A Visual Toolkit for Prompt Engineering and LLM Hypothesis Testing. In Proceedings of the CHI Conference on Human Factors in Computing Systems . 1–18

    work page 2024

  6. [6]

    Lisa P Argyle, Ethan C Busby, Nancy Fulda, Joshua R Gubler, Christopher Rytting, and David Wingate. 2023. Out of one, many: Using language models to simulate human samples. Political Analysis 31, 3 (2023), 337–351

    work page 2023

  7. [7]

    J Scott Armstrong, William B Denniston Jr, and Matt M Gordon. 1975. The use of the decomposition principle in making judgments. Organizational behavior and human performance 14, 2 (1975), 257–263

    work page 1975

  8. [8]

    Amanda Askell, Yuntao Bai, Anna Chen, Dawn Drain, Deep Ganguli, Tom Henighan, Andy Jones, Nicholas Joseph, Ben Mann, Nova DasSarma, et al. 2021. A general language assistant as a laboratory for alignment. arXiv preprint arXiv:2112.00861 (2021)

    arXiv 2021

Show all 69 references
  1. [9]

    Bharathi Asokarajan, Ronak Etemadpour, June Abbas, Samuel J Huskey, and Chris E Weaver. 2017. TexTile: A Pixel-Based Focus+ Context Tool For Analyzing Variants Across Multiple Text Scales.. In EuroVis (Short Papers). 49–53

    2017

  2. [10]

    Yuntao Bai, Andy Jones, Kamal Ndousse, Amanda Askell, Anna Chen, Nova DasSarma, Dawn Drain, Stanislav Fort, Deep Ganguli, Tom Henighan, et al. 2022. Training a helpful and harmless assistant with reinforcement learning from human feedback. arXiv preprint arXiv:2204.05862 (2022)

    2022 arXiv

  3. [11]

    Ralf Biedert, Georg Buscher, Sven Schwarz, Jörn Hees, and Andreas Dengel. 2010. Text 2.0. In CHI ’10 Extended Abstracts on Human Factors in Computing Systems (Atlanta, Georgia, USA) (CHI EA ’10) . Association for Computing Machinery, New York, NY, USA, 4003–4008. doi:10.1145/1...

    2010

  4. [12]

    Steven Bird. 2006. NLTK: the natural language toolkit. In Proceedings of the COLING/ACL 2006 interactive presentation sessions . 69–72

    2006

  5. [13]

    Michael Bostock, Vadim Ogievetsky, and Jeffrey Heer. 2011. D 3 data-driven documents. IEEE transactions on visualization and computer graphics 17, 12 (2011), 2301–2309

    2011

  6. [14]

    Stuart K Card, Jock Mackinlay, and Ben Shneiderman. 1999. Readings in informa- tion visualization: using vision to think . Morgan Kaufmann

    1999

  7. [15]

    Tuhin Chakrabarty, Philippe Laban, and Chien-Sheng Wu. 2024. Can AI writing be salvaged? Mitigating Idiosyncrasies and Improving Human-AI Alignment in the Writing Process through Edits. arXiv preprint arXiv:2409.14509 (2024)

    2024 arXiv

  8. [16]

    Guiming Hardy Chen, Shunian Chen, Ziche Liu, Feng Jiang, and Benyou Wang

  9. [17]

    Furui Cheng, Vilém Zouhar, Simran Arora, Mrinmaya Sachan, Hendrik Strobelt, and Mennatallah El-Assady. 2024. Relic: Investigating large language model responses using self-consistency. In Proceedings of the CHI Conference on Human Factors in Computing Systems . 1–18

    2024

  10. [18]

    Paul F Christiano, Jan Leike, Tom Brown, Miljan Martic, Shane Legg, and Dario Amodei. 2017. Deep reinforcement learning from human preferences. Advances in neural information processing systems 30 (2017)

    2017

  11. [19]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human...

    2019

  12. [20]

    Danica Dillion, Niket Tandon, Yuling Gu, and Kurt Gray. 2023. Can AI language models replace human participants? Trends in Cognitive Sciences 27, 7 (2023), 597–600

    2023

  13. [21]

    Edward W Dolch. 1936. A basic sight vocabulary. The Elementary School Journal 36, 6 (1936), 456–460

    1936

  14. [22]

    Mennatallah El-Assady, Rita Sevastjanova, Fabian Sperrle, Daniel Keim, and Christopher Collins. 2017. Progressive learning of topic modeling parameters: A visual analytics framework. IEEE transactions on visualization and computer graphics 24, 1 (2017), 382–391

    2017

  15. [23]

    Mennatallah El-Assady, Fabian Sperrle, Rita Sevastjanova, Michael Sedlmair, and Daniel Keim. 2018. LTMA: Layered topic matching for the comparative explo- ration, evaluation, and refinement of topic modeling results. In 2018 International Symposium on Big Data Visual and Immer...

    2018

  16. [24]

    Angela Fan, Aleksandra Piktus, Fabio Petroni, Guillaume Wenzek, Marzieh Saeidi, Andreas Vlachos, Antoine Bordes, and Sebastian Riedel. 2020. Generating Fact Checking Briefs. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) , Bon...

    2020

  17. [25]

    Simret Araya Gebreegziabher, Elena L Glassman, and Toby Jia-Jun Li. 2024. MOCHA: Model Optimization through Collaborative Human-AI Alignment. In Adjunct Proceedings of the 37th Annual ACM Symposium on User Interface Software and Technology. 1–4

    2024

  18. [26]

    Simret Araya Gebreegziabher, Yukun Yang, Elena L Glassman, and Toby Jia-Jun Li. 2024. Supporting Co-Adaptive Machine Teaching through Human Concept Learning and Cognitive Theories. arXiv preprint arXiv:2409.16561 (2024)

    2024 arXiv

  19. [27]

    Katy Ilonka Gero, Chelse Swoopes, Ziwei Gu, Jonathan K Kummerfeld, and Elena L Glassman. 2024. Supporting sensemaking of large language model outputs at scale. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems. 1–21

    2024

  20. [28]

    Jiawei Gu, Xuhui Jiang, Zhichao Shi, Hexiang Tan, Xuehao Zhai, Chengjin Xu, Wei Li, Yinghan Shen, Shengjie Ma, Honghao Liu, et al . 2024. A survey on llm-as-a-judge. arXiv preprint arXiv:2411.15594 (2024)

    2024 arXiv

  21. [29]

    Ziwei Gu, Ian Arawjo, Kenneth Li, Jonathan K Kummerfeld, and Elena L Glassman

  22. [30]

    Ziwei Gu, Owen Raymond, Naser Al Madi, and Elena L Glassman. 2024. Why Do Skimmers Perform Better with Grammar-Preserving Text Saliency Modulation (GP-TSM)? Evidence from an Eye Tracking Study. InExtended Abstracts of the CHI Conference on Human Factors in Computing Systems . 1–8

    2024

  23. [31]

    Dylan Hadfield-Menell, Stuart J Russell, Pieter Abbeel, and Anca Dragan. 2016. Cooperative inverse reinforcement learning. Advances in neural information processing systems 29 (2016)

    2016

  24. [32]

    In Proceedings of the CHI Conference on Human Factors in Computing Systems

    An AI-Resilient Text Rendering Technique for Reading and Skimming Documents. In Proceedings of the CHI Conference on Human Factors in Computing Systems. 1–22

  25. [33]

    Stefan Jänicke and David Joseph Wrisley. 2017. Interactive visual alignment of medieval text versions. In 2017 IEEE Conference on Visual Analytics Science and Technology (V AST). IEEE, 127–138

    2017

  26. [34]

    Nikhita Joshi and Daniel Vogel. 2024. Constrained Highlighting in a Document Reader can Improve Reading Comprehension. InProceedings of the CHI Conference on Human Factors in Computing Systems . 1–10

    2024

  27. [35]

    Franziska Huth, Maurice Koch, Miriam Awad-Mohammed, Daniel Weiskopf, and Kuno Kurzhals. 2024. Eye Tracking on Text Reading with Visual Enhancements. In Proceedings of the 2024 Symposium on Eye Tracking Research and Applications . 1–7

    2024

  28. [36]

    Tae Soo Kim, Yoonjoo Lee, Jamin Shin, Young-Ho Kim, and Juho Kim. 2024. Evallm: Interactive evaluation of large language model prompts on user-defined criteria. In Proceedings of the CHI Conference on Human Factors in Computing Systems. 1–21

    2024

  29. [37]

    Aniket Kittur, Ed H Chi, and Bongwon Suh. 2008. Crowdsourcing user studies with Mechanical Turk. In Proceedings of the SIGCHI conference on human factors in computing systems. 453–456

    2008

  30. [38]

    Minsuk Kahng, Ian Tenney, Mahima Pushkarna, Michael Xieyang Liu, James Wexler, Emily Reif, Krystal Kallarackal, Minsuk Chang, Michael Terry, and Lucas Dixon. 2024. LLM Comparator: Interactive Analysis of Side-by-Side Evaluation of Large Language Models. IEEE Transactions on Vi...

    2024

  31. [39]

    Cassidy Laidlaw and Anca Dragan. 2021. The Boltzmann policy distribution: Accounting for systematic suboptimality in human models. In International Conference on Learning Representations

    2021

  32. [40]

    Jan Leike, David Krueger, Tom Everitt, Miljan Martic, Vishal Maini, and Shane Legg. 2018. Scalable agent alignment via reward modeling: a research direction. arXiv preprint arXiv:1811.07871 (2018)

    2018 arXiv

  33. [41]

    Jan Kompatscher, Danqing Shi, Giovanna Varni, Tino Weinkauf, and Antti Oulasvirta. 2025. Interactive groupwise comparison for reinforcement learning from human feedback. arXiv preprint arXiv:2507.04340 (2025)

    2025

  34. [42]

    Androniki Mertsiotaki, Stephanie Hofmann, Sarah Keck, Emily Kratsch, Alexan- der Daum, and Birgit Popp. 2025. Designing Usable Interfaces for Human Eval- uation of LLM-Generated Texts: UX Challenges and Solutions. In International Joint Conference on Artificial Intelligence 2025

    2025

  35. [43]

    Yannick Metz, Andras Geiszl, Raphaël Baur, and Mennatallah El-Assady. 2025. Reward Learning from Multiple Feedback Types. In International Conference on Learning Representations

    2025

  36. [44]

    Yang Liu, Dan Iter, Yichong Xu, Shuohang Wang, Ruochen Xu, and Chenguang Zhu. 2023. G-eval: NLG evaluation using gpt-4 with better human alignment. arXiv preprint arXiv:2303.16634 (2023)

    2023 arXiv

  37. [45]

    Sewon Min, Kalpesh Krishna, Xinxi Lyu, Mike Lewis, Wen-tau Yih, Pang Koh, Mohit Iyyer, Luke Zettlemoyer, and Hannaneh Hajishirzi. 2023. FActScore: Fine- grained Atomic Evaluation of Factual Precision in Long Form Text Generation. In Proceedings of the 2023 Conference on Empiri...

    2023 doi

  38. [46]

    Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A Rusu, Joel Veness, Marc G Bellemare, Alex Graves, Martin Riedmiller, Andreas K Fidjeland, Georg Ostrovski, et al. 2015. Human-level control through deep reinforcement learning. nature 518, 7540 (2015), 529–533

    2015

  39. [47]

    Yannick Metz, David Lindner, Raphaël Baur, Daniel Keim, and Mennatallah El- Assady. 2023. RLHF-Blender: A Configurable Interactive Interface for Learning from Diverse Human Feedback. In ICML2023 Interactive Learning from Implicit Human Feedback Workshop

    2023

  40. [48]

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. 2022. Training language models to follow instructions with human feedback. Advances in neural information processing systems 35...

    2022

  41. [49]

    Ryan Park, Rafael Rafailov, Stefano Ermon, and Chelsea Finn. 2024. Disen- tangling length from quality in direct preference optimization. arXiv preprint arXiv:2403.19159 (2024)

    2024 arXiv

  42. [50]

    Ani Nenkova and Rebecca J Passonneau. 2004. Evaluating content selection in summarization: The pyramid method. In Proceedings of the human language tech- nology conference of the north american chapter of the association for computational linguistics: Hlt-naacl 2004. 145–152

    2004

  43. [51]

    Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. arXiv preprint arXiv:1908.10084 (2019)

    2019 arXiv

  44. [52]

    Rita Sevastjanova, Simon Vogelbacher, Andreas Spitz, Daniel Keim, and Menna- tallah El-Assady. 2023. Visual comparison of text sequences generated by large language models. In 2023 IEEE Visualization in Data Science (VDS) . IEEE, 11–20

    2023

  45. [53]

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. 2023. Direct preference optimization: Your language model is secretly a reward model. Advances in Neural Information Processing Systems 36 (2023), 53728–53741

    2023

  46. [54]

    Danqing Shi, Shibei Zhu, Tino Weinkauf, and Antti Oulasvirta. 2024. Interac- tive Reward Tuning: Interactive Visualization for Preference Elicitation. In 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 9254–9261

    2024

  47. [55]

    Nisan Stiennon, Long Ouyang, Jeffrey Wu, Daniel Ziegler, Ryan Lowe, Chelsea Voss, Alec Radford, Dario Amodei, and Paul F Christiano. 2020. Learning to summarize with human feedback. Advances in neural information processing systems 33 (2020), 3008–3021

    2020

  48. [56]

    Shreya Shankar, JD Zamfirescu-Pereira, Björn Hartmann, Aditya Parameswaran, and Ian Arawjo. 2024. Who validates the validators? aligning llm-assisted evalu- ation of llm outputs with human preferences. In Proceedings of the 37th Annual ACM Symposium on User Interface Software ...

    2024

  49. [57]

    Nicole Sultanum, Devin Singh, Michael Brudno, and Fanny Chevalier. 2018. Doc- curate: A curation-based approach for clinical text visualization.IEEE transactions on visualization and computer graphics 25, 1 (2018), 142–151

    2018

  50. [58]

    Richard S Sutton, Andrew G Barto, et al. 1998. Reinforcement learning: An intro- duction. Vol. 1. MIT press Cambridge. DxHF: Providing High-Quality Human Feedback for LLM Alignment UIST ’25, September 28-October 1, 2025, Busan, Republic of Korea

    1998

  51. [59]

    Hendrik Strobelt, Daniela Oelke, Bum Chul Kwon, Tobias Schreck, and Hanspeter Pfister. 2015. Guidelines for effective usage of text highlighting techniques. IEEE transactions on visualization and computer graphics 22, 1 (2015), 489–498

    2015

  52. [60]

    Wayne A Wickelgren. 1977. Speed-accuracy tradeoff and information processing dynamics. Acta psychologica 41, 1 (1977), 67–85

    1977

  53. [61]

    Robert F Woolson. 2005. Wilcoxon signed-rank test. Encyclopedia of biostatistics 8 (2005)

    2005

  54. [62]

    Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann Dubois, Xuechen Li, Carlos Guestrin, Percy Liang, and Tatsunori B Hashimoto. 2023. Stanford alpaca: An instruction-following llama model

    2023

  55. [63]

    Tariq Yousef and Stefan Janicke. 2020. A survey of text alignment visualization. IEEE transactions on visualization and computer graphics 27, 2 (2020), 1149–1159

    2020

  56. [64]

    Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric Xing, et al. 2023. Judging llm-as-a-judge with mt-bench and chatbot arena. Advances in Neural Information Processing Systems 36 (2023), 46595–46623

    2023

  57. [65]

    Dustin Wright, David Wadden, Kyle Lo, Bailey Kuehl, Arman Cohan, Isabelle Augenstein, and Lucy Lu Wang. 2022. Generating Scientific Claims for Zero-Shot Scientific Fact Checking. In Proceedings of the 60th Annual Meeting of the Associa- tion for Computational Linguistics (Volu...

    2022 doi

  58. [66]

    Given the conversation as the context: {Query}. Here are two claims: {Claim 1} and {Claim 2}. Please summarize the two claims. No other words

    Richard P Zipoli Jr. 2017. Unraveling difficult sentences: Strategies to support reading comprehension. Intervention in School and Clinic 52, 4 (2017), 218–227. UIST ’25, September 28-October 1, 2025, Busan, Republic of Korea Shi et al. A Prompts A.1 Decomposition Prompt We br...

    2017

  59. [68]

    Brian D Ziebart, J Andrew Bagnell, and Anind K Dey. 2010. Modeling interaction via the principle of maximum causal entropy. In ICML ’10: Proceedings of the 27th International Conference on Machine Learning . ACM, 1255–1262

    2010

  60. [2024]

    In Proceedings UIST ’25, September 28-October 1, 2025, Busan, Republic of Korea Shi et al

    Humans or LLMs as the Judge? A Study on Judgement Bias. In Proceedings UIST ’25, September 28-October 1, 2025, Busan, Republic of Korea Shi et al. of the 2024 Conference on Empirical Methods in Natural Language Processing , Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung Chen (Ed...

    2025 doi

  61. [7161]

    doi:10.18653/v1/2020.emnlp-main.580

    2020 doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.