Pith. sign in

REVIEW 5 major objections 4 minor 98 references

Generative UI tools fail to implement roughly a quarter of the design reasoning they state, with functional requirements failing at 34%.

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

2026-08-04 01:29 UTC pith:M45CQDXT

load-bearing objection Design Theater is a useful, benchmarkable construct; the 25% figure is plausible but the extraction step and partial-credit wording need validation before it counts as a stable measurement. the 5 major comments →

arxiv 2607.22928 v2 pith:M45CQDXT submitted 2026-07-24 cs.AI

Design Theater: Evaluating the Gap Between User-Facing Design Reasoning and Implementation in Generative UI Tools

Kashif Imteyaz , Kaif Imteyaz , Nakul Rajpal , Kaif Shaikh , Michael Muller , Saiph Savage This is my paper
classification cs.AI
keywords design theatergenerative UI toolsreasoning-to-implementation gapThinking Fidelity ScorePrinciple Adherence Scoredesign homogenizationfunctional requirementsUX principles
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.

Generative UI tools promise to let anyone type a description and get a finished interface, complete with a written rationale explaining the design. This paper argues that those rationales are often disconnected from what is actually built, a phenomenon the authors call Design Theater. Using 24 design tasks and five tools, they find that on average roughly 25% of a tool's stated design reasoning fails to appear in the delivered interface, and that the failure rises to 34% for functional requirements. They also find that tools recognize only about half of the UX principles embedded in prompts, and that different tools converge on similar visual appearance and layout for the same prompt. The point matters because non-expert users treat these rationales as design documentation, and may trust interfaces that do not work as described.

Core claim

The paper coins 'Design Theater' for the disconnect between the user-facing design reasoning a generative UI tool emits and the interface it actually implements. In a benchmark of 24 tasks across five tools (120 generated interfaces), it measures a mean Thinking Fidelity Score of 0.75, meaning that roughly one in four concrete, verifiable reasoning elements in the tool's own rationale is not fully realized in the generated output; for functional tasks the mean falls to 0.66, a 34% failure rate. Tools also implement on average only 54% of the two UX principles implicitly embedded in each prompt (PAS), and four of five tools implement 6% or fewer of the interaction-focused functional principle

What carries the argument

The argument rests on a benchmark of 24 natural-language UI tasks split into structural, styling, and functional tiers, plus three metrics: Thinking Fidelity Score (TFS), which scores each concrete, verifiable element in a tool's reasoning trace as fully implemented (1.0), partially (0.5), or not (0.0) against the rendered interface; Principle Adherence Score (PAS), which checks whether each of the two UX principles embedded in a prompt is implemented; and Design Homogeneity Index (DHI), which computes pairwise similarity of screenshots across tools via visual embedding, color histograms, and tree-edit distance on parsed layout. TFS carries the headline claim about the gap, PAS connects the

Load-bearing premise

The headline rates assume the benchmark's manual extraction of reasoning elements and its pre-assignment of two UX principles to each task are a valid measure of what each tool promised and delivered; the paper's own limitations note that tools were judged only against commitments they chose to state, under an enforced HTML/CSS/JS constraint, so different codebooks or conditions could shift the numbers.

What would settle it

Re-run the 24 tasks and have an independent team extract concrete claims from each tool's trace, then compare them to the rendered DOM with automated element matching. If the independent audit finds that nearly all extracted claims are implemented (TFS approaching 1.0), the Design Theater gap would shrink to a self-narration artifact; if the audit reproduces ~0.75, the finding is robust.

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

If this is right

  • Anyone who relies on a tool's rationale as documentation of the delivered interface will be misled about roughly a quarter of its stated design commitments, and about a third for functional behavior.
  • Functional failures—missing error recovery, user control, system status visibility—are the most common and the least detectable by static inspection, so generated interfaces require interactive testing.
  • Because different tools converge on similar appearance and layout for the same prompt, the ecosystem may flatten design variety even as rationales describe situated choices.
  • The TFS-PAS-DHI benchmark is a reusable audit: any generative UI tool can be scored on whether its reasoning matches its artifact, and whether its output contributes to homogenization.

Where Pith is reading between the lines

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

  • If the 25% gap generalizes, evaluation of generative UI tools should treat 'said but not built' as a first-class failure mode, on par with code that does not compile.
  • A plausible mechanism behind the functional gap is that tools optimize for visual surface plausibility while stateful behavior (clicks, errors, keyboard navigation) is harder to verify internally; a testable extension would measure whether tools improve when the prompt demands an interaction trace.
  • The DHI convergence finding suggests an intervention worth testing: injecting explicit diversity constraints or style anchors into prompts may reduce homogenization without reducing fidelity.
  • Because the paper counts only reasoning elements the tool chose to emit, the true gap could be larger; an independent audit that injects known requirements and checks the rendered DOM would provide a lower-bound test.

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

5 major / 4 minor

Summary. This paper introduces the Design Theater construct—a mismatch between user-facing design rationales produced by generative UI tools and the interfaces those tools actually implement. It contributes a 24-task benchmark across three tiers (structural, styling, functional), five generative UI tools, 120 generated interfaces, and three metrics: Thinking Fidelity Score (TFS), Principle Adherence Score (PAS), and Design Homogeneity Index (DHI). The headline findings are mean TFS 0.75, a functional-tier drop to 0.66, mean PAS 0.54 with near-zero functional principle adherence for most tools, and narrow pairwise distances in visual/layout similarity with larger color variation. The paper argues these results show that users who read tool rationales as descriptions of the delivered artifact will be misled on a substantive fraction of design commitments.

Significance. The work addresses a timely and understudied problem: the user-facing rationales emitted by generative UI tools are typically unverified and may induce overtrust. The public corpus, the explicit metric definitions, bootstrap confidence intervals, and inter-rater reliability for the Full/Partial/Not classification are genuine strengths. If the extraction and principle-mapping steps are validated, the benchmark and metrics could become a useful community resource for evaluating generative UI systems. However, the headline percentages currently overstate what the metrics actually measure, so the empirical contribution is real but the claims need recalibration.

major comments (5)
  1. [Abstract; Eq. (1)] Eq. (1) credits partially implemented elements with 0.5, so TFS = 0.75 means a quarter of the available credit is missing, not that 25% of stated rationales are 'not implemented.' If any element received 0.5, the frequency of zero-credit elements is strictly below 25%. The Abstract, Introduction, and Conclusion nevertheless say 'not implemented' or 'fails to appear,' while the Discussion more carefully says 'did not fully appear.' Report the not-implemented proportion separately from the credit-weighted score and revise the headline wording.
  2. [§Metric 1, steps 1–2] The extraction step in Metric 1 is the gate that determines Ntotal, and it is not reliability-checked. The paper reports weighted κ=0.70 only for the Full/Partial/Not classification (step 2), not for element extraction or for the 'concrete and verifiable' filter (step 1). No extraction codebook, procedure (manual vs. LLM-based), or excluded examples are provided. Because tools emit different proportions of vague versus concrete language, the cross-tool TFS differences and the 0.75 mean may reflect extractor judgment rather than tool behavior. Please add extraction-level reliability, make the codebook and extracted-element list public, and give examples of excluded statements.
  3. [§Metric 2, Scoring Protocol] PAS is defined relative to the authors' a priori assignment of exactly two target UX principles to each of 24 tasks. The manuscript does not list the 48 principle-task assignments or provide reliability or validation for that mapping. The phrase 'implicitly embedded' is not operationalized; different coders could plausibly assign different principles to the same prompt, which would change every PAS value in Table 2 and the headline mean of 0.54. Provide the full mapping, a written selection protocol, and ideally second-coder agreement on the mapping itself.
  4. [§Metric 3; Table 3] The abstract and Discussion conclude that designs 'converge' in visual appearance and layout, but the DHI sub-measures are raw distances with no reference distribution: DHI-Visual 0.119–0.151, DHI-Layout 0.181–0.211, DHI-Color 20.6–39.7. The paper acknowledges this lack in Limitations, yet the convergence claim is presented as a finding. Without a human-designed baseline or a random-pair/null distribution, these values cannot support 'convergence'; they can only support 'similarity among the five tools in this sample.' Add a baseline or soften the claim.
  5. [§Prompting and Generative UI Output Collection] The fixed implementation instruction (only HTML/CSS/JS, no external frameworks, libraries, or web searches) may depress functional-task performance relative to the tools' default behavior. The 34% functional TFS gap and the Tier-3 PAS ≤0.06 are the paper's most consequential findings, but the abstract reports them without this constraint. TFS is less affected because it is scored against the tool's own claims, but PAS and DHI are directly affected. Acknowledge this constraint in the abstract or provide a robustness condition using tools' default stacks.
minor comments (4)
  1. [Figure 1] Figure 1 appears to show only point estimates; adding confidence intervals or error bars would align it with Table 1 and make the cross-tool uncertainty visible.
  2. [Appendix / Data Availability] Include the 24 prompts and the PAS principle mapping in an appendix. The GitHub link is useful, but the benchmark should be self-contained in the paper for reproducibility.
  3. [References] Several references are future-dated or have inconsistent formatting (e.g., Anthropic 2026, W3C 2024 entries). Standardize and, where applicable, add arXiv IDs or DOIs.
  4. [Results, PAS section] The statement that non-overlapping 95% bootstrap CIs provide 'descriptive evidence' is appropriate, but the paper should avoid implying formal significance tests were run.

Circularity Check

0 steps flagged

No circular derivation: TFS, PAS, and DHI are descriptive measurements against external artifacts/principles; self-citations are not load-bearing.

full rationale

The paper's central quantities are empirical measurements, not derived predictions. TFS compares extracted concrete reasoning elements to the rendered interface (Eq. 1); the observed mean of 0.75 and Tier-3 mean of 0.66 could have been different, so the finding is not forced by the metric's definition. The extraction step's exclusion of vague statements narrows the construct but does not entail the measured gap. PAS uses externally grounded HCI principles assigned before evaluation, and the near-zero functional scores could plausibly have been high, so low PAS is not encoded in the benchmark design. DHI relies on external models (UIClip, CIELCh histograms, OmniParser) and pairwise distances; no fitted parameter is renamed as a prediction. The self-citation to Imteyaz et al. 2026 appears only as a related-work example and is not load-bearing. The paper's own Limitations section acknowledges scope constraints—'tools were evaluated only against commitments they chose to make' and DHI 'lacks a human-designed reference distribution'—which are construct-validity caveats, not circular steps. The abstract's phrase 'not implemented' is stronger than the partial-credit equation implies, but that is a reporting/framing issue rather than a reduction of the result to its inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 5 axioms · 1 invented entities

No numeric free parameters or fitted constants appear in the metrics; the ledger burden is conceptual: the benchmark, the two-principles-per-task mapping, the reasoning-element extraction rule, and the DHI similarity models are author-selected assumptions rather than independently verified instruments.

axioms (5)
  • domain assumption Human evaluators can reliably classify natural-language reasoning elements into Fully/Partially/Not implemented against rendered interfaces.
    TFS computation depends entirely on this; the paper reports weighted κ=0.70, which is substantial but leaves room for annotation variance; the extraction step itself has no reported reliability.
  • ad hoc to paper Each of the 24 benchmark prompts has exactly two right implicit UX principles, determinable before evaluation.
    The framework-first selection is authored for this benchmark; PAS divides by 2, so the score assumes the principle set is exactly two per task (Eq. 2).
  • domain assumption Excluding 'vague' reasoning elements does not bias TFS.
    Metric 1 excludes statements like 'make it user-friendly,' but the boundary between vague and concrete is subjective and could differentially affect tools that narrate more abstractly.
  • domain assumption Default tool configurations and one generation per task per tool represent the tool's behavior.
    The study captures one snapshot; the paper acknowledges outputs may vary by model version and repeated generations (Limitations).
  • domain assumption UIClip, CIELCh EMD, and OmniParser tree-edit distance are valid similarity proxies for homogenization.
    DHI sub-measures rest on these models; no human baseline or validation against perceived similarity is provided (Limitations).
invented entities (1)
  • Design Theater construct no independent evidence
    purpose: Names and operationalizes the rationale-implementation gap in generative UI tools
    The construct is defined by the paper and measured by metrics built for it; no external validation (e.g., correlation with user trust/error rates or expert judgment) is provided.

pith-pipeline@v1.3.0-alltime-deepseek · 18952 in / 14335 out tokens · 140222 ms · 2026-08-04T01:29:59.272357+00:00 · methodology

0 comments
read the original abstract

Generative UI tools promise to democratize UI design by turning natural language descriptions into complete interfaces. Alongside the interface, these tools generate user-facing design rationales that explain their layout, accessibility, and design choices. However, it remains unclear whether these stated rationales are actually reflected in the interfaces they produce. We call this disconnect ``Design Theater'': plausible and confident design rationales that have little relationship to the actual implementation. To study this phenomenon, we introduce a benchmark and three metrics for measuring Design Theater. The benchmark includes 24 UI generation tasks spanning structural, styling, and functional design requirements. Using this benchmark, we evaluate 120 interfaces created by five generative UI tools. On average, over 25\% of user-facing design rationales are not implemented in the generated interface, and the implementation failure increases to 34\% for functional requirements. Tools recognize roughly half of the UX principles embedded in prompts (mean = 0.54), with four of five tools implementing 6\% or fewer functional principles. We also measure interface similarity across tools and find convergence in visual appearance and layout organization, with greater variation in color choices. Overall, we contribute: 1) the concept of Design Theater; 2) a benchmark with metrics for assessing whether the stated reasoning of generative UI tools is reflected in their implementations; 3) and findings from a systematic evaluation of these tools. We discuss what these findings mean for the design and evaluation of generative UI tools.

Figures

Figures reproduced from arXiv: 2607.22928 by Kaif Imteyaz, Kaif Shaikh, Kashif Imteyaz, Michael Muller, Nakul Rajpal, Saiph Savage.

Figure 1
Figure 1. Figure 1: Overall Thinking Fidelity Score by Tool Tool Overall Tier 1 Tier 2 Tier 3 Structural Styling Functional Claude 0.87 [0.82, 0.91] 0.83 [0.74, 0.91] 0.93 [0.89, 0.96] 0.85 [0.78, 0.91] Bolt 0.84 [0.75, 0.91] 0.83 [0.70, 0.95] 0.95 [0.90, 0.99] 0.73 [0.55, 0.89] ChatGPT 0.80 [0.75, 0.84] 0.85 [0.75, 0.94] 0.78 [0.72, 0.85] 0.75 [0.69, 0.81] Vercel v0 0.74 [0.63, 0.84] 0.83 [0.70, 0.94] 0.83 [0.66, 0.96] 0.56 … view at source ↗
Figure 2
Figure 2. Figure 2: Thinking Fidelity Score across Tier, by Tools [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Principle Adherence Score across five UI agent tools. Left: mean PAS with 95% confidence intervals. Right: PAS by [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Design Homogeneity Index Heat Map the expertise to judge whether it reflects their goals, satis￾fies requirements, or follows sound UX principles. The chal￾lenge becomes more consequential when generative UI tools also provide user-facing explanations of what was designed and why (Sun et al. 2026). By articulating layout decisions, invoking design principles, and describing tradeoffs in the language of tra… view at source ↗

Review history (2 revisions) →

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

98 extracted references · 7 linked inside Pith

  1. [1]

    2021 , howpublished=

    GitHub Copilot: Your AI Pair Programmer , author=. 2021 , howpublished=

    2021

  2. [2]

    2024 , howpublished=

    axe-core: Accessibility Testing Engine , author=. 2024 , howpublished=

    2024

  3. [3]

    , title =

    Replit, Inc. , title =. 2026 , url =

    2026

  4. [4]

    , title =

    Vercel Inc. , title =. 2026 , url =

    2026

  5. [5]

    2025 , month =

    Lovable Dev , title =. 2025 , month =

    2025

  6. [6]

    2024 , howpublished =

    Google Chrome , title =. 2024 , howpublished =

    2024

  7. [7]

    2025 , month=

    Codex , author=. 2025 , month=

    2025

  8. [8]

    Proceedings of the 37th Annual ACM Symposium on User Interface Software and Technology , pages=

    UIClip: a data-driven model for assessing user interface design , author=. Proceedings of the 37th Annual ACM Symposium on User Interface Software and Technology , pages=

  9. [9]

    2015 , publisher=

    Information architecture: for the web and beyond , author=. 2015 , publisher=

    2015

  10. [10]

    arXiv preprint arXiv:2408.00203 , year=

    Omniparser for pure vision based gui agent , author=. arXiv preprint arXiv:2408.00203 , year=

    Pith/arXiv arXiv

  11. [11]

    2025 , howpublished=

    Claude Code: Agentic Coding System , author=. 2025 , howpublished=

    2025

  12. [12]

    2024 , month = dec, day =

    Web Content Accessibility Guidelines (WCAG) 2.2 , author =. 2024 , month = dec, day =

    2024

  13. [13]

    2013 , publisher=

    The design of everyday things: Revised and expanded edition , author=. 2013 , publisher=

    2013

  14. [14]

    Proceedings of the 2024 ACM Designing Interactive Systems Conference , pages=

    Generative AI in user experience design and research: how do UX practitioners, teams, and companies use GenAI in industry? , author=. Proceedings of the 2024 ACM Designing Interactive Systems Conference , pages=

    2024

  15. [15]

    Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems , pages=

    Co-Designing Collaborative Generative AI Tools for Freelancers , author=. Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems , pages=

    2026

  16. [16]

    2011 , publisher=

    Design research through practice: From the lab, field, and showroom , author=. 2011 , publisher=

    2011

  17. [17]

    Proceedings of the SIGCHI Conference on Human Factors in Computing Systems , pages=

    Reading critical designs: supporting reasoned interpretations of critical design , author=. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems , pages=

  18. [18]

    Proceedings of the 2020 CHI conference on human factors in computing systems , pages=

    Interpreting interpretability: understanding data scientists' use of interpretability tools for machine learning , author=. Proceedings of the 2020 CHI conference on human factors in computing systems , pages=

    2020

  19. [19]

    Human--Computer Interaction , pages=

    Exploring the application of LLM-based AI in UX design: an empirical case study of ChatGPT , author=. Human--Computer Interaction , pages=. 2024 , publisher=

    2024

  20. [20]

    Proceedings of the 2024 CHI conference on human factors in computing systems , pages=

    User experience design professionals’ perceptions of generative artificial intelligence , author=. Proceedings of the 2024 CHI conference on human factors in computing systems , pages=

    2024

  21. [21]

    Announcing v0: Generative UI , year =

  22. [22]

    Introducing Canvas , year =

  23. [23]

    What Are Artifacts and How Do I Use Them? , year =

  24. [24]

    Firebase Studio , year =

  25. [25]

    Introduction to Bolt , year =

  26. [26]

    Proceedings of the 2022 ACM conference on fairness, accountability, and transparency , pages=

    Designing for responsible trust in AI systems: A communication perspective , author=. Proceedings of the 2022 ACM conference on fairness, accountability, and transparency , pages=

    2022

  27. [27]

    Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems , pages=

    ClearFairy: Capturing Creative Workflows through Decision Structuring, In-Situ Questioning, and Rationale Inference , author=. Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems , pages=

    2026

  28. [28]

    Proceedings of the 2021 CHI conference on human factors in computing systems , pages=

    Investigating the homogenization of web design: A mixed-methods approach , author=. Proceedings of the 2021 CHI conference on human factors in computing systems , pages=

    2021

  29. [29]

    2024 , howpublished=

    Cursor: The AI Code Editor , author=. 2024 , howpublished=

    2024

  30. [30]

    Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education

    Clancey, William J. Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education. Proceedings of the Eighth International Joint Conference on Artificial Intelligence (IJCAI-83)

  31. [31]

    Classification Problem Solving

    Clancey, William J. Classification Problem Solving. Proceedings of the Fourth National Conference on Artificial Intelligence

  32. [32]

    , title =

    Robinson, Arthur L. , title =. 1980 , doi =. https://science.sciencemag.org/content/208/4447/1019.full.pdf , journal =

    1980

  33. [33]

    New Ways to Make Microcircuits Smaller---Duplicate Entry

    Robinson, Arthur L. New Ways to Make Microcircuits Smaller---Duplicate Entry. Science

  34. [34]

    Clancey and Glenn Rennels , abstract =

    Diane Warner Hasling and William J. Clancey and Glenn Rennels , abstract =. Strategic explanations for a diagnostic consultation system , journal =. 1984 , issn =. doi:https://doi.org/10.1016/S0020-7373(84)80003-6 , url =

    doi:10.1016/s0020-7373(84)80003-6 1984

  35. [35]

    and Rennels, Glenn R

    Hasling, Diane Warner and Clancey, William J. and Rennels, Glenn R. and Test, Thomas. Strategic Explanations in Consultation---Duplicate. The International Journal of Man-Machine Studies

  36. [36]

    Poligon: A System for Parallel Problem Solving

    Rice, James. Poligon: A System for Parallel Problem Solving

  37. [37]

    Transfer of Rule-Based Expertise through a Tutorial Dialogue

    Clancey, William J. Transfer of Rule-Based Expertise through a Tutorial Dialogue

  38. [38]

    The Engineering of Qualitative Models

    Clancey, William J. The Engineering of Qualitative Models

  39. [39]

    2017 , eprint=

    Attention Is All You Need , author=. 2017 , eprint=

    2017

  40. [40]

    Pluto: The 'Other' Red Planet

    NASA. Pluto: The 'Other' Red Planet

  41. [41]

    arXiv preprint arXiv:2304.10778 , year=

    Evaluating the code quality of ai-assisted code generation tools: An empirical study on github copilot, amazon codewhisperer, and chatgpt , author=. arXiv preprint arXiv:2304.10778 , year=

    Pith/arXiv arXiv

  42. [42]

    Participatory design , pages=

    Scandinavian design: On participation and skill , author=. Participatory design , pages=. 2017 , publisher=

    2017

  43. [43]

    Participatory design , pages=

    Contextual inquiry: A participatory technique for system design , author=. Participatory design , pages=. 2017 , publisher=

    2017

  44. [44]

    Conference companion on Human factors in computing systems , pages=

    Usability inspection methods , author=. Conference companion on Human factors in computing systems , pages=

  45. [45]

    Adjunct Proceedings of the 31st Annual ACM Symposium on User Interface Software and Technology , pages =

    Oulasvirta, Antti and De Pascale, Samuli and Koch, Janin and Langerak, Thomas and Jokinen, Jussi and Todi, Kashyap and Laine, Markku and Kristhombuge, Manoj and Zhu, Yuxi and Miniukovich, Aliaksei and Palmas, Gregorio and Weinkauf, Tino , title =. Adjunct Proceedings of the 31st Annual ACM Symposium on User Interface Software and Technology , pages =. 201...

    arXiv 2018

  46. [46]

    Proceedings of the Extended Abstracts of the 2026 CHI Conference on Human Factors in Computing Systems , pages=

    Seeing the Reasoning: How LLM Rationales Influence User Trust and Decision-Making in Factual Verification Tasks , author=. Proceedings of the Extended Abstracts of the 2026 CHI Conference on Human Factors in Computing Systems , pages=

    2026

  47. [47]

    Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems , articleno =

    Chung, John Joon Young and Kim, Wooseok and Yoo, Kang Min and Lee, Hwaran and Adar, Eytan and Chang, Minsuk , title =. Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems , articleno =. 2022 , isbn =. doi:10.1145/3491102.3501819 , abstract =

    arXiv 2022

  48. [48]

    arXiv preprint arXiv:2405.07065 , volume=

    Logomotion: Visually grounded code generation for content-aware animation , author=. arXiv preprint arXiv:2405.07065 , volume=

    Pith/arXiv arXiv

  49. [49]

    arXiv preprint arXiv:2402.06071 , year=

    Keyframer: Empowering animation design using large language models , author=. arXiv preprint arXiv:2402.06071 , year=

    Pith/arXiv arXiv

  50. [50]

    arXiv preprint arXiv:1910.08930 , year=

    Sketch2Code: transformation of sketches to UI in real-time using deep neural network , author=. arXiv preprint arXiv:1910.08930 , year=

    Pith/arXiv arXiv 1910

  51. [51]

    Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems , pages=

    DesignWeaver: dimensional scaffolding for text-to-image product design , author=. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems , pages=

    2025

  52. [52]

    Proceedings of the ACM SIGCHI Symposium on Engineering Interactive Computing Systems , articleno =

    Beltramelli, Tony , title =. Proceedings of the ACM SIGCHI Symposium on Engineering Interactive Computing Systems , articleno =. 2018 , isbn =. doi:10.1145/3220134.3220135 , abstract =

    arXiv 2018

  53. [53]

    arXiv preprint arXiv:2307.13702 , year=

    Measuring faithfulness in chain-of-thought reasoning , author=. arXiv preprint arXiv:2307.13702 , year=

    Pith/arXiv arXiv

  54. [54]

    2025 , eprint=

    Vibe coding: programming through conversation with artificial intelligence , author=. 2025 , eprint=

    2025

  55. [55]

    Proceedings of the 5th Annual Symposium on Human-Computer Interaction for Work , pages=

    Vibe Coding in Product Teams: Reconfiguring AI-Assisted Workflows, Prototyping, and Collaboration , author=. Proceedings of the 5th Annual Symposium on Human-Computer Interaction for Work , pages=

  56. [56]

    Proceedings of the 16th conference on creativity & cognition , pages=

    Homogenization effects of large language models on human creative ideation , author=. Proceedings of the 16th conference on creativity & cognition , pages=

  57. [57]

    Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems , articleno =

    Agarwal, Dhruv and Naaman, Mor and Vashistha, Aditya , title =. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems , articleno =. 2025 , isbn =. doi:10.1145/3706598.3713564 , abstract =

    arXiv 2025

  58. [58]

    Proceedings of the 2024 CHI conference on human factors in computing systems , pages=

    The effects of generative AI on design fixation and divergent thinking , author=. Proceedings of the 2024 CHI conference on human factors in computing systems , pages=

    2024

  59. [59]

    Science advances , volume=

    Generative AI enhances individual creativity but reduces the collective diversity of novel content , author=. Science advances , volume=. 2024 , publisher=

    2024

  60. [60]

    Communications of the ACM , volume=

    Asleep at the keyboard? assessing the security of github copilot’s code contributions , author=. Communications of the ACM , volume=. 2025 , publisher=

    2025

  61. [61]

    2025 , eprint=

    Vibe Coding in Practice: Motivations, Challenges, and a Future Outlook -- a Grey Literature Review , author=. 2025 , eprint=

    2025

  62. [62]

    2025 , issue_date =

    Fu, Yujia and Liang, Peng and Tahir, Amjed and Li, Zengyang and Shahin, Mojtaba and Yu, Jiaxin and Chen, Jinfu , title =. 2025 , issue_date =. doi:10.1145/3716848 , journal =

    doi:10.1145/3716848 2025

  63. [63]

    Advances in neural information processing systems , volume=

    Chain-of-thought prompting elicits reasoning in large language models , author=. Advances in neural information processing systems , volume=

  64. [64]

    Advances in Neural Information Processing Systems , volume=

    Language models don't always say what they think: Unfaithful explanations in chain-of-thought prompting , author=. Advances in Neural Information Processing Systems , volume=

  65. [65]

    Proceedings of the human factors and ergonomics society annual meeting , volume=

    NASA-task load index (NASA-TLX); 20 years later , author=. Proceedings of the human factors and ergonomics society annual meeting , volume=. 2006 , organization=

    2006

  66. [66]

    Usability evaluation in industry , volume=

    SUS-A quick and dirty usability scale , author=. Usability evaluation in industry , volume=. 1996 , publisher=

    1996

  67. [67]

    2021 , publisher=

    Engineering psychology and human performance , author=. 2021 , publisher=

    2021

  68. [68]

    2020 , publisher=

    Design justice: Community-led practices to build the worlds we need , author=. 2020 , publisher=

    2020

  69. [69]

    Women, science, and technology , pages=

    Situated knowledges: The science question in feminism and the privilege of partial perspective 1 , author=. Women, science, and technology , pages=. 2013 , publisher=

    2013

  70. [70]

    Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

    Gaps in the safety evaluation of generative AI , author=. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

  71. [71]

    Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

    Trustworthy social bias measurement , author=. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

  72. [72]

    Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

    Can we trust ai benchmarks? an interdisciplinary review of current issues in ai evaluation , author=. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

  73. [73]

    Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

    Towards interactive evaluations for interaction harms in human-AI systems , author=. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

  74. [74]

    ACM computing surveys , volume=

    Survey of hallucination in natural language generation , author=. ACM computing surveys , volume=. 2023 , publisher=

    2023

  75. [75]

    2021 , isbn =

    Wu, Jason and Zhang, Xiaoyi and Nichols, Jeff and Bigham, Jeffrey P , title =. 2021 , isbn =. doi:10.1145/3472749.3474763 , booktitle =

    arXiv 2021

  76. [76]

    2023 , isbn =

    Wu, Jason and Krosnick, Rebecca and Schoop, Eldon and Swearngin, Amanda and Bigham, Jeffrey P and Nichols, Jeffrey , title =. 2023 , isbn =. doi:10.1145/3586183.3606824 , booktitle =

    arXiv 2023

  77. [77]

    and Rezazadegan Tavakoli, Hamed and R

    Jiang, Yue and Leiva, Luis A. and Rezazadegan Tavakoli, Hamed and R. B. Houssel, Paul and Kylm\". UEyes: Understanding Visual Saliency across User Interface Types , year =. doi:10.1145/3544548.3581096 , booktitle =

    arXiv

  78. [78]

    International journal of qualitative methods , volume=

    Intercoder reliability in qualitative research: Debates and practical guidelines , author=. International journal of qualitative methods , volume=. 2020 , publisher=

    2020

  79. [79]

    Proceedings of the ACM on human-computer interaction , volume=

    Reliability and inter-rater reliability in qualitative research: Norms and guidelines for CSCW and HCI practice , author=. Proceedings of the ACM on human-computer interaction , volume=. 2019 , publisher=

    2019

  80. [80]

    International journal of computer vision , volume=

    The earth mover's distance as a metric for image retrieval , author=. International journal of computer vision , volume=. 2000 , publisher=

    2000

Showing first 80 references.