Pith. sign in

REVIEW 3 major objections 6 minor 61 references

Animated transitions from prompt to AI response help users locate, verify, and trust generated content, with measured improvements of 43%, 153%, and 20%.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 15:11 UTC pith:3TLA3DTP

load-bearing objection A solid HCI contribution — a useful taxonomy and well-run studies — but the broad applicability claim rests on mapping reliability the paper hasn't yet shown. the 3 major comments →

arxiv 2607.18507 v1 pith:3TLA3DTP submitted 2026-07-20 cs.HC

AInimation: Animating from Prompt to AI-Generated Responses

Jiaqi Wu , Damien Masson This is my paper
classification cs.HC
keywords animated transitionshuman-AI interactionprompt engineeringresponse reviewmultimodal generative AIuser studytaxonomyexplainability
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper argues that the default 'instant' or token-by-token display of AI responses is a missed opportunity: by animating the mapping between a user's prompt and the generated response, users can better track where their input took effect, see what the AI changed, and judge whether the AI understood the request. The authors build a taxonomy of seven animation types (reuse, alteration, reconceptualization, structural, internal reference, external reference, extraneous) and show in three user experiments that these animations outperform a same-duration instant baseline. If correct, the work implies that AI interfaces should spend some of their speed budget on meaningful animation rather than pure immediacy.

Core claim

The central discovery is that a well-designed animated transition, which visually connects elements of the prompt to their counterparts in the AI response, materially improves a user's ability to review and trust that response. Across three experiments with 16 participants, animated transitions led to a 43% reduction in error when locating referenced elements in the response, a 152% reduction in error when estimating added/removed content during modification tasks, and a 20% increase in confidence that the AI correctly interpreted a prompt instruction. The authors establish a taxonomy of seven prompt-response relationship types—reuse, alteration, reconceptualization, structural, internal ref

What carries the argument

The central mechanism is the 'animated transition' built upon a causal element mapping between the prompt and the response, defined by the counterfactual condition that removing an element from the prompt would remove the corresponding element in the response. The taxonomy classifies seven types of this relationship and prescribes a corresponding animation pattern (e.g., translation via morphing, red/green flash for alterations, overlay-and-fade for structural requirements, side-by-side resolution for external references). The timing framework coordinates dependent animations and uses 'slow-in slow-out' temporal distortion to aid tracking.

Load-bearing premise

The taxonomy and all the tested animations presuppose that meaningful causal mappings between prompt elements and response elements can be reliably identified—yet in practice the automated pipeline achieves only 0.62 accuracy for alteration and 0.59 for external reference, and the taxonomy itself was created by a single reviewer without inter-rater reliability.

What would settle it

A replication study using a larger, more diverse participant pool (e.g., older adults, non-English speakers) that compares animated vs. instant conditions on real tasks (not just the laboratory proxy tasks) would falsify the claim if the performance benefits shrink to non-significance. Additionally, a deployment where the automatic pipeline supplies the mappings (rather than hand-crafted ones) would test whether the 43%/153%/20% improvements hold despite the pipeline's imperfect accuracy.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If AI systems adopt these animations, users will spend less effort locating information in generated answers, potentially reducing the cost of verifying AI output.
  • For editing tools (e.g., grammar correction, code refactoring), animated alteration highlighting could make AI modifications transparent and auditable, improving trust.
  • The 'slower but well-crafted' animation approach challenges the current design dogma that faster generation is always better, suggesting a UX trade-off worth revisiting.
  • The taxonomy provides a common vocabulary and reusable animation specs for multimodal AI interfaces (text, image, text-to-image, image-to-text).
  • The automatic pipeline's imperfect accuracy (especially for alteration and external reference) implies that near-term practical deployments should combine automatic detection with human-in-the-loop or fallback strategies.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The measured benefits are likely conservative because the study compared animation against a baseline where users had equal extra time to study the final static response; real-world 'instant' displays often cut off as soon as generation ends, so the gap could be larger in practice.
  • The taxonomy's 'internal reference' and 'external reference' animations highlight the interpretability of the model's resolution process; if these become standard, they could serve as lightweight explainability tools for end users without requiring model introspection.
  • A testable extension is to adapt the 'alteration' animation for code diffs in AI pair-programming tools, where locating and understanding AI-introduced changes is a known pain point.
  • The single-author taxonomy (no inter-rater reliability reported) is a validity threat that a follow-up with multiple annotators could address; if the taxonomy fails to generalize beyond the sampled 800 pairs, the animation benefits may not transfer.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes AInimation, a set of animated transitions between a user prompt and an AI-generated response, grounded in a seven-category taxonomy (reuse, alteration, reconceptualization, structural, internal reference, external reference, extraneous) derived from 800 prompt–response pairs. The authors implement a prototype player and a naive automatic pipeline for generating animation specifications, then report three user experiments comparing animated transitions with an equal-time instant-response baseline. The headline findings are a 43% improvement in locating elements, a 152–153% improvement in estimating changes, and a 20% improvement in confidence that the prompt was correctly interpreted, with generally positive subjective ratings. The paper concludes that well-crafted, slower animations should replace instant AI responses across AI-integrated software.

Significance. If the central claim holds, this is a useful and timely HCI contribution: it offers an output-communication strategy for AI systems that is temporary, does not consume permanent screen real estate, and could be applied to chat interfaces, image editing, and code assistants. The paper has clear strengths: the taxonomy is organized around concrete design goals (G1–G4); the user experiments use matched equal-time baselines, pre-planned analysis, studentized bootstrap confidence intervals, and honest reporting of the inconclusive external-reference sub-result; and the limitations section explicitly acknowledges weaknesses in the technical implementation. However, the significance depends on two load-bearing assumptions that are not yet established: that the prompt-to-response element mappings required by every animation can be produced reliably in real deployments, and that the observed benefits come from animation per se rather than from the explicit mapping information that animation happens to convey.

major comments (3)
  1. [§4.2, Table 1; §5; §7] The paper's broad applicability claim — 'our work applies to all software that integrates AI' — presupposes that the mapping and animation-type decisions can be automated. But Section 4.2 reports only 0.62 accuracy for alteration and 0.59 for external reference (Table 1), precisely the categories that most require semantic interpretation. The user studies in Section 5 use animations 'generated from our prototype system,' but the animation specifications in those studies are not shown to come from the automatic pipeline; in fact the pipeline evaluation is presented separately and the system requires hand-provided masks and external-reference resolutions. Thus the experiments demonstrate an upper bound: when the correct mappings are known in advance, animation helps. Real deployments with automatically inferred mappings will include missed or spurious animations, and the headline benefits
  2. [§5, Experiments 1–3] The experimental design compares animated transitions against an instant-response baseline, but the animated condition differs from the baseline in two ways simultaneously: it contains motion and it contains explicit visual correspondence information (which prompt element maps to which response element, which words were added/deleted, which area of the image changed). The paper therefore cannot attribute the obtained benefits to animation per se; a static color-coded overlay or a static before/after side-by-side view might confer the same benefits, or most of them. Given the paper's conclusion that 'well-crafted, slower animations are preferable to instant AI responses,' it is important to include a control condition with static correspondence highlighting but no motion. Without this control, the argument that animation specifically (rather than explicit mapping information) is what help
  3. [§3.1–§3.2] The taxonomy and the element mappings on which every animation depends were developed by a single author reviewing 800 prompt–response pairs, with no inter-rater reliability or second-labeler validation. The mapping definition in Section 3.2 is counterfactual ('removing the element from the prompt would remove the corresponding element in the response'), but this paper does not operationalize how that counterfactual is assessed in practice, nor does it report how reliably different annotators would agree on the mapping or on the animation category. This matters because the validation step ('we sampled another 100 pairs... and verified that each fit within our taxonomy') was presumably performed by the same author(s), not by independent raters. I am not asking for a full IRR study, but some evidence of inter-rater agreement on the taxonomy and on a sample of element mappings would strengt
minor comments (6)
  1. [§5.3.1] In the confidence results, the fourth modality is listed as 'text-to-image' again; it should be 'text-to-text' (the four modalities are text-to-text, text-to-image, image-to-text, image-to-image).
  2. [Abstract vs §5.3.1] The abstract reports '153% better at identifying changes', while Section 5.3.1 reports a '152% improvement'; please make these consistent.
  3. [§4.1] Typo: 'boostrapped' should be 'bootstrapped'. Also, '10,0000 resamples' should be '10,000 resamples'.
  4. [§6.1] The summary says participants were '25% more confident at locating elements' and '44% more confident' at estimating changes, but the results are reported as raw mean differences on a 5-point scale (0.62 and 1.08 points). Translating these into percentages without defining the denominator is misleading; please either report raw differences in the summary or define the percentage conversion.
  5. [§5.5.1] For 'The animation was engaging' and 'The animation was easy to understand', the paper reports 'Mdn=4 95% [4.09 4.57]'. A median cannot have a 95% CI that lies entirely above 4 when the median is reported as 4; this appears to be a mean that was mislabeled as a median, or the CI brackets are misreported.
  6. [Figure 2] The taxonomy figure is extremely dense and difficult to read, especially the image-to-image rows. A larger version or a supplementary high-resolution figure would help readers verify the animation types and their cross-modal variants.

Circularity Check

0 steps flagged

No circular derivation: the taxonomy is dataset-derived, the pipeline is evaluated against it, and the user studies compare hand-authored animations against an independent instant baseline.

full rationale

The paper is an empirical study rather than a derivation, and I found no step in which a claimed result reduces by construction to its own inputs. The taxonomy (Section 3.3) is presented as the outcome of reviewing 800 prompt–response pairs (Section 3.1), not as a prediction derived from the experiments or from the pipeline. The technical feasibility study (Section 4) evaluates the automatic pipeline by manually checking whether its outputs follow the taxonomy (Section 4.1: 'We evaluated the accuracy of these generated animation specifications by manually labelling them and checking whether the produced animations follow the descriptions in our taxonomy.'); that is a benchmark-style evaluation of a separate component, not a circular proof that the taxonomy is correct. The user studies (Section 5) test animations produced by the prototype with known element mappings against an instant baseline with matched viewing time; they therefore support the stated conclusion about well-crafted animations without presupposing the pipeline's imperfect accuracy. The paper itself flags the key limitation that automatic mapping is imperfect (Table 1: alteration 0.62, external reference 0.59) and explicitly frames the pipeline as 'the floor of performance one can expect' — this is a validity/scope caveat, not circularity. Self-citations (Textoshop, DirectGPT, Statslator) appear only as related work or application examples and are not load-bearing for the central claim. The mild observation that experiment prompts were 'taking inspiration from real prompts found during our analysis' is a stimulus-design choice, not a reduction of the measured effect to the taxonomy; the effect was measured against an independent baseline. No fitted parameter is renamed as a prediction, and no uniqueness theorem or ansatz is imported from prior work to force the conclusion. Hence no specific circular step can be quoted, and the appropriate score is 0.

Axiom & Free-Parameter Ledger

1 free parameters · 5 axioms · 0 invented entities

The central claim is empirical, not derived, so the ledger records domain assumptions rather than fitted quantities. No numbers are fitted to data; the only hand-set values are animation timings taken from prior work. The load-bearing premises are the feasibility of element mapping, the single-annotator taxonomy, the instant baseline as the correct comparator, and the transferability of lab tasks. The paper discloses most of these in Section 6.3, which lowers the circularity burden but not the uncertainty.

free parameters (1)
  • Animation durations (reuse 1.5s, reconceptualization 2s, structural/external-reference overlay timing, total 4-6s)
    Hand-set in Section 3.4.3 from prior work and the authors' own tests, not fitted to the study data. They bound the equal-time comparison and the headline claim that 'slower animations are preferable.'
axioms (5)
  • domain assumption Prompt and response elements are related by a causal relationship: removing the prompt element would remove the corresponding response element.
    Section 3.2 defines mappings this way. The counterfactual is not verifiable per pair and grounds all seven animation types.
  • domain assumption A single author's review of 800 prompt-response pairs yields a comprehensive, unbiased taxonomy.
    Section 3.1: one reviewer highlighted all elements and generalized animations; validation used another 100 pairs per dataset but no second coder or inter-rater reliability statistic.
  • domain assumption Lab tasks are valid proxies for real-world AI response review.
    Stated in Section 5 and acknowledged in Section 6.3.2. Tasks are simplified (clicking locations, counting added/removed elements, rating confidence), and in Experiment 3 all responses actually followed instructions.
  • domain assumption Instant presentation is the strongest baseline for comparison.
    Section 5: word-by-word and diffusion-like alternatives were pilot-tested and deemed weak, so the equal-time instant baseline was assumed to be the fairest comparator.
  • domain assumption The proprietary models used (GPT-5.4, ChatGPT-5.2-thinking, Nano Banana 2) produce outputs stable enough for the reported pipeline and stimulus results to replicate.
    Sections 4.1 and 5: the pipeline uses an undisclosed 'optimized prompt'; results will vary across model versions and runs, so replication depends on unavailable artifacts.

pith-pipeline@v1.3.0-alltime-deepseek · 18850 in / 17246 out tokens · 182217 ms · 2026-08-01T15:11:18.502049+00:00 · methodology

0 comments
read the original abstract

We explore the use of animated transitions between a prompt and an AI-generated response. After reviewing 800 examples of prompts and responses, we devise a taxonomy of animated transitions for multimodal text- and image-generative models. The proposed animations include translating and morphing elements of the prompt to their final location in the response; highlighting modifications such as fixed typos; overlaying structural requirements to verify them; and displaying how a model understands references. A study shows that adding animated transitions helps users review the response: participants performed 43% better at locating elements in the response; 153% better at identifying changes; and 20% better at verifying the prompt was correctly interpreted. Our work applies to all software that integrates AI and shows that well-crafted, slower animations are preferable to instant AI responses.

Figures

Figures reproduced from arXiv: 2607.18507 by Damien Masson, Jiaqi Wu.

Figure 1
Figure 1. Figure 1: We design, demonstrate, and evaluate a set of animations between a prompt and a response: (a) elements of the image [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: reuse animation for text to image generation. The last step shows the beginning of extraneous animation. Alteration. Elements from the prompt are reused but with slight modifications. A frequent case is when asking an LLM to “fix the 3 [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 2
Figure 2. Figure 2: Our complete taxonomy of animated transitions across considered modalities and types of relationship between [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 5
Figure 5. Figure 5: Example reconceptualization animation. The last step shows the beginning of extraneous animation. Structural. Elements from the prompt decide the structure and format of the response. For example, a prompt might ask for a specific length or a specific shape (e.g., bullet points). Similarly, when generating or modifying an image, a prompt might provide relational instructions (e.g., “not bigger than the dog… view at source ↗
Figure 6
Figure 6. Figure 6: structural animation for text-to-text generation. The structural overlay will eventually fade out. Internal Reference. Elements from the prompt refer to other el￾ements from the prompt. For example, pronouns such as “he” or “these” refer to previously mentioned names or expressions. Some of these pronouns might be an ambiguous reference. We suggest highlighting the referee and the referent in the prompt wi… view at source ↗
Figure 7
Figure 7. Figure 7: external reference animation. Prompt elements "mona lisa" and "style of Wassily Kandinsky" morph into images and then move and dissolve into the response image. Extraneous. For all elements in the response that do not have a clear relationship to the prompt, we propose to show them through a fade-in animation at the very end. For example, LLMs such as ChatGPT tend to add extra information, such as encourag… view at source ↗
Figure 9
Figure 9. Figure 9: Average error in pixels between participants’ clicks [PITH_FULL_IMAGE:figures/full_fig_p007_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Absolute error when estimating changes. Each dot [PITH_FULL_IMAGE:figures/full_fig_p008_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Responses to likert statements during the three experiments for (a) animated transitions and (b) baseline (instant [PITH_FULL_IMAGE:figures/full_fig_p009_11.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

61 extracted references · 5 canonical work pages

  1. [1]

    [n. d.]. Inception – A New Frontier in LLM Speed. https://www.inceptionlabs.ai/

  2. [2]

    Samira Abnar and Willem Zuidema. 2020. Quantifying Attention Flow in Trans- formers. InProceedings of the 58th Annual Meeting of the Association for Compu- tational Linguistics. 4190–4197

    2020

  3. [3]

    Pepa Atanasova, Sarthak Jain, Sweta Porada, and Byron C. Wallace. 2020. A Diagnostic Study of Explainability Methods for Text Classification. InProceedings of the 2020 Conference on Empirical Methods in Natural Language Processing. 3635–3652

    2020

  4. [4]

    Jinbin Bai, Wei Chow, Ling Yang, Xiangtai Li, Juncheng Li, Hanwang Zhang, and Shuicheng Yan. 2024. Humanedit: A high-quality human-rewarded dataset for instruction-based image editing.arXiv preprint arXiv:2412.04280(2024)

    Pith/arXiv arXiv 2024

  5. [5]

    Patrick Baudisch, Desney Tan, Maxime Collomb, Dan Robbins, Ken Hinckley, Maneesh Agrawala, Shengdong Zhao, and Gonzalo Ramos. 2006. Phosphor: Ex- plaining Transitions in the User Interface Using Afterglow Effects. InProceedings of the 19th Annual ACM Symposium on User Interface Software and Technology (UIST ’06). Association for Computing Machinery, New Y...

    arXiv 2006

  6. [6]

    Bederson and A

    B.B. Bederson and A. Boltman. 1999. Does Animation Help Users Build Mental Maps of Spatial Information?. InProceedings 1999 IEEE Symposium on Information Visualization (InfoVis’99). 28–35. doi:10.1109/INFVIS.1999.801854

    arXiv 1999

  7. [7]

    Maalvika Bhat. 2025. How Dynamic vs. Static Presentation Shapes User Percep- tion and Emotional Connection to Text-Based AI. InProceedings of the 30th Inter- national Conference on Intelligent User Interfaces (IUI ’25). Association for Com- puting Machinery, New York, NY, USA, 846–860. doi:10.1145/3708359.3712131

    arXiv 2025

  8. [8]

    Aaron Chatterji, Tom Cunningham, David Deming, Zoe Hitzig, Christopher Ong, Carl Shan, and Kevin Wadman. 2025. How People Use ChatGPT. (2025)

    2025

  9. [10]

    Fanny Chevalier, Pierre Dragicevic, Anastasia Bezerianos, and Jean-Daniel Fekete

  10. [11]

    Fanny Chevalier, Pierre Dragicevic, and Steven Franconeri. 2014. The Not-so- Staggering Effect of Staggered Animated Transitions on Visual Tracking.IEEE Transactions on Visualization and Computer Graphics20, 12 (Dec. 2014), 2241–2250. doi:10.1109/TVCG.2014.2346424

    arXiv 2014

  11. [12]

    InProceedings of the SIGCHI Conference on Human Factors in Computing Systems

    Using text animated transitions to support navigation in document histories. InProceedings of the SIGCHI Conference on Human Factors in Computing Systems. 683–692

  12. [13]

    Christopher Chou, Lisa Dunlap, Koki Mashita, Krishna Mandal, Trevor Darrell, Ion Stoica, Joseph E Gonzalez, and Wei-Lin Chiang. 2025. Visionarena: 230k real world user-vlm conversations with preference labels. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 3877–3887

    2025

  13. [14]

    Fanny Chevalier, Nathalie Henry Riche, Catherine Plaisant, Amira Chalbi, and Christophe Hurter. 2016. Animations 25 Years Later: New Roles and Opportuni- ties. InProceedings of the International Working Conference on Advanced Visual Interfaces (A VI ’16). Association for Computing Machinery, New York, NY, USA, 280–287. doi:10.1145/2909132.2909255

    arXiv 2016

  14. [15]

    2016.Introduction to the New Statistics: Estimation, Open Science, and Beyond(1st edition ed.)

    Geoff Cumming and Robert Calin-Jageman. 2016.Introduction to the New Statistics: Estimation, Open Science, and Beyond(1st edition ed.). Routledge, London : New York

    2016

  15. [16]

    Chun and Yuhong Jiang

    Marvin M. Chun and Yuhong Jiang. 1998. Contextual Cueing: Implicit Learning and Memory of Visual Context Guides Spatial Attention.Cognitive Psychology 36, 1 (June 1998), 28–71. doi:10.1006/cogp.1998.0681

    arXiv 1998

  16. [17]

    Pierre Dragicevic, Anastasia Bezerianos, Waqas Javed, Niklas Elmqvist, and Jean-Daniel Fekete. 2011. Temporal Distortion for Animated Transitions. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI ’11). Association for Computing Machinery, New York, NY, USA, 2009–2018. doi:10.1145/1978942.1979233

    arXiv 2011

  17. [18]

    Pierre Dragicevic. 2016. Fair Statistical Communication in HCI. InModern Statistical Methods for HCI, Judy Robertson and Maurits Kaptein (Eds.). Springer International Publishing, Cham, 291–330. doi:10.1007/978-3-319-26633-6_13

    doi:10.1007/978-3-319-26633-6_13 2016

  18. [19]

    Pierre Dragicevic, Stéphane Huot, and Fanny Chevalier. 2011. Gliimpse: Ani- mating from markup code to rendered documents and vice versa. InProceedings of the 24th annual ACM symposium on User interface software and technology. 257–262

    2011

  19. [20]

    Pierre Dragicevic, Stéphane Huot, and Fanny Chevalier. 2011. Gliimpse: Animat- ing from Markup Code to Rendered Documents and Vice Versa. InProceedings of the 24th Annual ACM Symposium on User Interface Software and Technology (UIST ’11). Association for Computing Machinery, New York, NY, USA, 257–262. doi:10.1145/2047196.2047229

    arXiv 2011

  20. [21]

    Shi Feng, Eric Wallace, Alvin Grissom II, and Zhaofeng Ren. 2018. Patholo- gies of Neural Models Make Interpretations Difficult. InProceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. 3262–3273

    2018

  21. [22]

    Fan Du, Nan Cao, Jian Zhao, and Yu-Ru Lin. 2015. Trajectory Bundling for Animated Transitions. InProceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems (CHI ’15). Association for Computing Machinery, New York, NY, USA, 289–298. doi:10.1145/2702123.2702476

    arXiv 2015

  22. [23]

    Jeffrey Heer and George Robertson. 2007. Animated Transitions in Statistical Data Graphics.IEEE Transactions on Visualization and Computer Graphics13, 6 (Nov. 2007), 1240–1247. doi:10.1109/TVCG.2007.70539

    arXiv 2007

  23. [24]

    Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde- Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio

    Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde- Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. 2014. Generative Adversarial Nets. InAdvances in Neural Information Processing Systems, Vol. 27. Curran Associates, Inc

    2014

  24. [25]

    Hudson and John T

    Scott E. Hudson and John T. Stasko. 1993. Animation Support in a User Interface Toolkit: Flexible, Robust, and Reusable Abstractions. InProceedings of the 6th Annual ACM Symposium on User Interface Software and Technology (UIST ’93). Association for Computing Machinery, New York, NY, USA, 57–67. doi:10.1145/ 168642.168648

    arXiv 1993

  25. [26]

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. 2020. Denoising Diffusion Probabilistic Models. InAdvances in Neural Information Processing Systems, Vol. 33. Curran Associates, Inc., 6840–6851

    2020

  26. [28]

    Sarthak Jain and Byron C. Wallace. 2019. Attention is not Explanation. InPro- ceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 3503–3514

    2019

  27. [29]

    Bederson

    Christian Klein and Benjamin B. Bederson. 2005. Benefits of Animated Scrolling. InCHI ’05 Extended Abstracts on Human Factors in Computing Systems (CHI EA ’05). Association for Computing Machinery, New York, NY, USA, 1965–1968. doi:10.1145/1056808.1057068

    arXiv 2005

  28. [30]

    Kingma and Max Welling

    Diederik P. Kingma and Max Welling. 2022. Auto-Encoding Variational Bayes. arXiv:1312.6114 [stat] doi:10.48550/arXiv.1312.6114

    Pith/arXiv arXiv doi:10.48550/arxiv.1312.6114 2022

  29. [31]

    Duan Li, Xinyuan Guo, Xinhuan Shu, Lanxi Xiao, Lingyun Yu, and Shixia Liu. 2025. RouteFlow: Trajectory-Aware Animated Transitions. InProceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI ’25). Association for Computing Machinery, New York, NY, USA, 1–17. doi:10.1145/3706598.3714300

    arXiv 2025

  30. [32]

    Hao-Ping (Hank) Lee, Advait Sarkar, Lev Tankelevitch, Ian Drosos, Sean Rintel, Richard Banks, and Nicholas Wilson. 2025. The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers. InProceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CH...

    arXiv 2025

  31. [33]

    Hashimoto

    Xiang Lisa Li, John Thickstun, Ishaan Gulrajani, Percy Liang, and Tatsunori B. Hashimoto. 2022. Diffusion-LM Improves Controllable Text Generation. InPro- ceedings of the 36th International Conference on Neural Information Processing Systems (NIPS ’22). Curran Associates Inc., Red Hook, NY, USA, 4328–4343

    2022

  32. [34]

    Jiwei Li, Will Monroe, and Dan Jurafsky. 2016. Understanding Neural Networks through Representation Erasure. InProceedings of the 2016 Conference on Empirical Methods in Natural Language Processing. 1405–1414

    2016

  33. [35]

    Lovelace and Stephen D

    Eugene A. Lovelace and Stephen D. Southall. 1983. Memory for Words in Prose and Their Locations on the Page.Memory & Cognition11, 5 (Sept. 1983), 429–434. doi:10.3758/BF03196979

    work page doi:10.3758/bf03196979 1983

  34. [36]

    Zhuoyan Li, Chen Liang, Jing Peng, and Ming Yin. 2024. The Value, Benefits, and Concerns of Generative AI-Powered Assistance in Writing. InProceedings of the 2024 CHI Conference on Human Factors in Computing Systems (CHI ’24). Association for Computing Machinery, New York, NY, USA, 1–25. doi:10.1145/ 3613904.3642625

    arXiv 2024

  35. [37]

    Damien Masson, Young-Ho Kim, and Fanny Chevalier. 2025. Textoshop: In- teractions Inspired by Drawing Software to Facilitate Text Editing. InPro- ceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI ’25). Association for Computing Machinery, New York, NY, USA, 1–14. doi:10.1145/3706598.3713862

    arXiv 2025

  36. [38]

    Chilana, Andrea Bunt, and Joanna McGrenere

    Shareen Mahmud, Jessalyn Alvina, Parmit K. Chilana, Andrea Bunt, and Joanna McGrenere. 2020. Learning Through Exploration: How Children, Adults, and Older Adults Interact with a New Feature-Rich Application. InProceedings of the 2020 CHI Conference on Human Factors in Computing Systems. ACM, Honolulu HI USA, 1–14. doi:10.1145/3313831.3376414 11 , , Jiaqi ...

    arXiv 2020

  37. [41]

    2014.How People Read on the Web: The Eyetracking Evidence

    Kara Pernice, Kathryn Whitenton, and Jakob Nielsen. 2014.How People Read on the Web: The Eyetracking Evidence. Nielsen Norman Group

    2014

  38. [42]

    Tomáš Mikolov, Martin Karafiát, Lukáš Burget, Jan Černocký, and Sanjeev Khu- danpur. 2010. Recurrent Neural Network Based Language Model. InProc. Inter- speech 2010. 1045–1048. doi:10.21437/Interspeech.2010-343

    doi:10.21437/interspeech.2010-343 2010

  39. [43]

    George Robertson, Kim Cameron, Mary Czerwinski, and Daniel Robbins. 2002. Animated Visualization of Multiple Intersecting Hierarchies.Information Visual- ization1, 1 (March 2002), 50–65. doi:10.1057/palgrave.ivs.9500002

    work page doi:10.1057/palgrave.ivs.9500002 2002

  40. [44]

    Alec Radford and Karthik Narasimhan. 2018. Improving Language Understanding by Generative Pre-Training

    2018

  41. [45]

    Ronald Baecker and Ian Small. 1990. Animation at the Interface. InThe Art of Human-Computer Interface Design

    1990

  42. [46]

    Rogers and Stephen Monsell

    Robert D. Rogers and Stephen Monsell. 1995. Costs of a Predictible Switch between Simple Cognitive Tasks.Journal of Experimental Psychology: General 124, 2 (1995), 207–231. doi:10.1037/0096-3445.124.2.207

    work page doi:10.1037/0096-3445.124.2.207 1995

  43. [47]

    Simons and Daniel T

    Daniel J. Simons and Daniel T. Levin. 1997. Change Blindness.Trends in Cognitive Sciences1, 7 (Oct. 1997), 261–267. doi:10.1016/S1364-6613(97)01080-2

    work page doi:10.1016/s1364-6613(97)01080-2 1997

  44. [48]

    Maruthappan Shanmugasundaram and Pourang Irani. 2008. The Effect of Ani- mated Transitions in Zooming Interfaces. InProceedings of the Working Conference on Advanced Visual Interfaces (A VI ’08). Association for Computing Machinery, New York, NY, USA, 396–399. doi:10.1145/1385569.1385642

    arXiv 2008

  45. [49]

    Sangho Suh, Bryan Min, Srishti Palani, and Haijun Xia. 2023. Sensecape: En- abling Multilevel Exploration and Sensemaking with Large Language Models. In Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology. ACM, San Francisco CA USA, 1–18. doi:10.1145/3586183.3606756

    arXiv 2023

  46. [50]

    Sangho Suh, Meng Chen, Bryan Min, Toby Jia-Jun Li, and Haijun Xia. 2024. Luminate: Structured Generation and Exploration of Design Space with Large Language Models for Human-AI Co-Creation. InProceedings of the CHI Conference on Human Factors in Computing Systems. ACM, Honolulu HI USA, 1–26. doi:10. 1145/3613904.3642400

    arXiv 2024

  47. [51]

    Mukund Sundararajan, Ankur Taly, and Qiqi Yan. 2017. Axiomatic Attribution for Deep Networks. InProceedings of the 34th International Conference on Machine Learning, Vol. 70. 3149–3157

    2017

  48. [52]

    Keqiang Sun, Junting Pan, Yuying Ge, Hao Li, Haodong Duan, Xiaoshi Wu, Renrui Zhang, Aojun Zhou, Zipeng Qin, Yi Wang, et al. 2023. Journeydb: A benchmark for generative image understanding.Advances in neural information processing systems36 (2023), 49659–49678

    2023

  49. [53]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Ł ukasz Kaiser, and Illia Polosukhin. 2017. Attention Is All You Need. InAdvances in Neural Information Processing Systems, Vol. 30. Curran Associates, Inc

    2017

  50. [54]

    Aäron van den Oord, Nal Kalchbrenner, and Koray Kavukcuoglu. 2016. Pixel Recurrent Neural Networks. InProceedings of The 33rd International Conference on Machine Learning. PMLR, 1747–1756

    2016

  51. [55]

    Scheidegger, and Huamin Qu

    Yong Wang, Daniel Archambault, Carlos E. Scheidegger, and Huamin Qu. 2018. A Vector Field Design Approach to Animated Transitions.IEEE Transactions on Visualization and Computer Graphics24, 9 (Sept. 2018), 2487–2500. doi:10.1109/ TVCG.2017.2750689

    arXiv 2018

  52. [56]

    Jesse Vig. 2019. BertViz: Visualizing Attention in Transformers. arXiv:1904.02679 [cs.CL]

    Pith/arXiv arXiv 2019

  53. [57]

    Chang Xiao and Zixiaofan Yang. 2025. Streaming, Fast and Slow: Cognitive Load- Aware Streaming for Efficient LLM Serving. InProceedings of the 38th Annual ACM Symposium on User Interface Software and Technology (UIST ’25). Association for Computing Machinery, New York, NY, USA, 1–13. doi:10.1145/3746059.3747721

    arXiv 2025

  54. [58]

    Sarah Wiegreffe and Yuval Pinter. 2019. Attention is not not Explanation. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Pro- cessing and the 9th International Joint Conference on Natural Language Processing. 11–20

    2019

  55. [59]

    Zamfirescu-Pereira, Richmond Y

    J.D. Zamfirescu-Pereira, Richmond Y. Wong, Bjoern Hartmann, and Qian Yang

  56. [60]

    Fisher, R

    Ka-Ping Yee, D. Fisher, R. Dhamija, and M. Hearst. 2001. Animated Exploration of Dynamic Graphs with Radial Layout. InIEEE Symposium on Information Visualization, 2001. INFOVIS 2001.43–50. doi:10.1109/INFVIS.2001.963279

    arXiv 2001

  57. [61]

    Yixian Zheng, Wenchao Wu, Nan Cao, Huamin Qu, and Lionel M. Ni. 2018. Focus+context Grouping for Animated Transitions.Journal of Visual Languages & Computing48 (Oct. 2018), 61–69. doi:10.1016/j.jvlc.2018.06.006

    work page doi:10.1016/j.jvlc.2018.06.006 2018

  58. [62]

    Yaqian Zhu and John Kolassa. 2018. Assessing and Comparing the Accuracy of Various Bootstrap Methods.Communications in Statistics - Simulation and Computation47, 8 (Sept. 2018), 2436–2453. doi:10.1080/03610918.2017.1348516 12

    arXiv 2018

  59. [63]

    Wenting Zhao, Xiang Ren, Jack Hessel, Claire Cardie, Yejin Choi, and Yuntian Deng. 2024. Wildchat: 1m chatgpt interaction logs in the wild.arXiv preprint arXiv:2405.01470(2024)

    Pith/arXiv arXiv 2024

  60. [2010]

    InProceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI ’10)

    Using Text Animated Transitions to Support Navigation in Document Histories. InProceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI ’10). Association for Computing Machinery, New York, NY, USA, 683–692. doi:10.1145/1753326.1753427

    arXiv

  61. [2023]

    InProceedings of the 2023 CHI Conference on Human Factors in Com- puting Systems

    Why Johnny Can’t Prompt: How Non-AI Experts Try (and Fail) to Design LLM Prompts. InProceedings of the 2023 CHI Conference on Human Factors in Com- puting Systems. ACM, Hamburg Germany, 1–21. doi:10.1145/3544548.3581388

    arXiv 2023