Pith. sign in

REVIEW 4 major objections 4 minor 47 references

Orchid: Orchestrating Context Across Creative Workflows with Generative AI

T0 review · 4 major / 4 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 A notebook that lets users specify, reference, and monitor context across a creative workflow produces more novel and feasible generative-AI outcomes than a chat-plus-search toolkit, according to a within-subjects study of 12 users.

desk verdict A well-described design probe whose evaluation supports the integrated system but not the specific causal claim about context orchestration; needs a more honest framing and some statistical cleanup before publication. read the letter →

arxiv 2508.19517 v1 pith:GFEPFRPQ submitted 2025-08-27 cs.HC cs.AI

Srishti Palani , Gonzalo Ramos This is my paper · ORCID
classification cs.HCcs.AI
keywords contextorchestrationgenerativeAIcreativitysupporttoolshuman-AIalignmentpersonasLLMpromptingnotebookinterfaceuserstudy
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

This paper argues that the main obstacle to useful generative-AI assistance in creative work is not model capability but context: users cannot easily tell the AI which project, personal, and stylistic information stays relevant as work evolves. To test that, it introduces Orchid, a notebook that lets users specify context through project pages, a personal Me page, and simulated expert personas; reference it explicitly with @ mentions, by selecting text inline, or by placing a prompt next to relevant content; and monitor what each prompt is grounded on through a transparency lens. In a within-subjects study of 12 participants, expert raters scored ideas produced with Orchid at $M=11.50$ out of 15 on overall creativity versus $M=6.45$ for a baseline of web search, chat, and a digital notebook, and users issued about half as many prompts. The paper takes this as evidence that orchestrated context improves alignment between user intent and AI output, and it argues that context management should be a first-class capability in next-generation creativity support tools.

What carries the argument

The central object is the meta-prompt: Orchid's back end takes the user's short prompt and expands it with whichever project documents, personal preferences, persona definitions, and goal or task breakdowns the user has referenced or implicitly grounded, then sends that assembled prompt to a large language model. The load-bearing interaction mechanisms are the @ mention for explicit references, in-line selection prompting, implicit grounding by page placement, and the Transparency Lens that shows what context a prompt is using. These mechanisms turn context into editable, named artifacts, such as project pages, a Me page, and Persona pages, that the user can reuse without retyping them in each prompt.

What would settle it

Run a follow-up study with the same integrated notebook in both arms, where the only difference is whether @ mentions, in-line grounding, personas, and the transparency lens are enabled; if expert-rated creativity and self-reported alignment are no better in the enabled arm, the paper's causal claim about context orchestration fails. A simpler falsifier: if a baseline chat tool that already supports @-style context attachments produces the same creativity scores as Orchid in a head-to-head comparison, the contribution reduces to interface packaging rather than context orchestration.

Watch

Extended reading notes

Core claim

The central claim is that context orchestration, giving users first-class affordances to specify, reference, and monitor context across sessions and models, is what lets generative AI act as a creative partner rather than a prompt-response engine. The evidence is the comparative study: blind expert raters judged outcomes more novel, feasible, and valuable under Orchid than under the baseline toolkit, and participants reported higher alignment with their goals, more control, more transparency, and a more empathetic, collaborative relationship with the AI. Behavioral logs show that Orchid users grounded operations in context frequently, with an average of 24.75 context references per participant, while baseline users switched tools about 15.75 times and copy-pasted about 8.56 times per session while issuing significantly more prompts. The conclusion is that reducing the cost of carrying context across iterations improves both the creative product and the experience of working with AI.

Load-bearing premise

The study assumes that the better outcomes come from the context-orchestration features, but the Orchid condition also differs from the baseline in being a single integrated notebook with task decomposition, persona templates, asynchronous generation, and a consistent interface, so the specific effect of context features is not isolated.

Editorial extensions

If this is right

  • If context orchestration is the active ingredient, chat-based assistants that expose editable, persistent context objects should outperform raw long-context windows on multi-session creative tasks.
  • Tools that let users see which context a prompt used should reduce the metacognitive load of validating AI output, because users can check whether the system had the right information before trusting the result.
  • Personas and personal context pages turn prompt engineering from a typing skill into a curation skill: users spend effort defining reusable lenses rather than restating them per prompt.
  • Designers of creativity support tools should treat context as a managed artifact with its own lifecycle, not as an implicit property of a single session.

Reading between the lines

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

  • Not an Orchid claim: the study varies several features together, so the causal role of context orchestration per se is not isolated; a fair reading is that Orchid as a whole beats the baseline toolkit, not that each affordance is necessary.
  • A natural extension the authors do not test is adding only the reference and transparency affordances to an ordinary chat tool, without the notebook shell, which would test whether context referencing alone drives the observed gains.
  • If the meta-prompt mechanism is the real driver, any interface that lowers the cost of attaching context, even a simple file-folder metaphor, should show similar creativity gains, which would predict that user control over what is in context matters more than context-window size.
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 / 4 minor

Summary. The paper presents Orchid, a GenAI-enabled notebook system that lets users specify, reference, and monitor context across creative workflows. Context specification is supported through project pages, a personal 'Me' page, and editable expert personas; referencing is supported through @-mentions, in-line selection, and implicit grounding; monitoring is supported through a Transparency Lens and result blocks. The paper reports a within-subjects user study (n=12) in which participants completed creative redesign tasks once with Orchid and once with a baseline stack of web search, LLM chat, and a notebook. Blind expert raters scored the creative outputs on novelty, feasibility, and value, and the paper reports a significantly higher combined creativity score for Orchid (M=11.50 vs 6.45, reported p=0.01), non-significantly higher novelty, significantly higher feasibility and value, lower prompt counts, and higher self-reported alignment, control, transparency, and empathetic support. The paper's central claim is that context orchestration, rather than raw context-window size or fragmented tools, can improve human-AI alignment and creative outcomes.

Significance. If the reported effect could be attributed specifically to context orchestration, the contribution would be useful to the creativity support tools and human-AI interaction communities: the design goals are clearly articulated, the system description is unusually complete, and the authors ship the actual prompt templates in the supplementary material. The within-subjects design, blind expert raters, counterbalanced order, and logging of interaction behavior are strengths. However, the evidence as presented does not isolate context orchestration from the many other differences between the Orchid condition and the baseline, and the abstract overstates the novelty result. The central design thesis is plausible and the empirical direction is worth reporting, but the causal attribution needs either additional analysis or a substantially more careful framing.

major comments (4)
  1. [§3, §4.2, §5.1] The comparison that supports the paper's central claim is confounded: the Orchid condition differs from the baseline not only in context orchestration but also in integration into a single notebook surface, goal decomposition and task pages, persona templates, asynchronous generation, and the transparency lens. Because the expert-rated outcome advantage in §5.1 could plausibly be caused by any of these features, the study as designed cannot attribute the effect to context orchestration specifically. I am not asserting the effect is absent, but the central claim of the abstract ('By prioritizing context orchestration...') requires an ablation, a feature-level mediation analysis, or a within-Orchid correlation between context-referencing behavior (e.g., @-mentions) and the expert-rated outcomes; alternatively, the claims should be reframed as a comparison of an integrated notebook versus a fragmented tool stack.
  2. [§5.1, Abstract] The abstract states that participants 'produced more novel and feasible outcomes,' but the novelty difference was not statistically significant (M=3.38 vs 1.92, reported as non-significant). The headline claim should be limited to the combined creativity measure (and feasibility/value), with novelty explicitly reported as a directional but non-significant trend. Additionally, the measures section does not report inter-rater reliability for the blind expert ratings (e.g., ICC or Cohen's kappa), and it is unclear whether the reported means refer to individual outputs or to a single per-participant aggregate; both details are needed to interpret the ratings.
  3. [§5.1–§5.2, Figure 11] The statistical reporting is not interpretable as written: 'P=1.58,Q=0.01↑↑' and 'WSRT: R=↓1.94,Q=0.03↑' do not name the test statistic or its distribution, do not report effect sizes, and do not state whether multiple-comparison corrections were applied. The notation 'N O=2.10' is also undefined (presumably standard deviation or standard error). Please provide standard test names, degrees of freedom, effect sizes, and define all symbols; otherwise readers cannot verify the significance claims.
  4. [§5.2, §5.3] The log-analysis claims that Orchid reduced prompts per session (M=12.00 vs 6.53) and increased context grounding (M=24.75 prompts) are presented as evidence for context orchestration, but these metrics are not normalized by the total number of actions or task complexity, and the baseline's tool-switching/copy-paste counts are not formally compared to Orchid behavior. A direct comparison of total interaction cost or a within-participant ratio would strengthen the claim that Orchid reduces redundant prompting.
minor comments (4)
  1. [§3.2] The heading 'Refering to Context' and numerous OCR artifacts in the text ('a!ordances', 'work"ows', 'Washignton') should be corrected in the camera-ready version.
  2. [§5.5] The quote attributed to P15 appears in a study that reports n=12; please verify the participant identifiers so the data are internally consistent.
  3. [§4.4] The procedure states that the presentation order of conditions was counterbalanced but does not state whether the two task topics were also counterbalanced with conditions; please clarify the topic-condition assignment to rule out topic effects.
  4. [Appendix A] The provided prompt templates contain formatting artifacts (e.g., '{{⁄quotesingle.Vartext}}') that make them difficult to reuse; please provide clean, copyable prompt source.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central evaluation is empirical and externally rated; the only self-citation is a design-motivation reference that is not load-bearing.

full rationale

The paper's central claim is that users produced better creative outcomes with Orchid than with a baseline toolkit. That claim rests on a within-subjects study with blind expert ratings of novelty, feasibility, and value (Section 5.1), behavioral log analyses (Section 5.2), and self-report survey measures (Sections 5.2-5.4). These outcome measures are not defined in terms of Orchid's context-orchestration features, no parameters are fitted to make the results come out, and the observed differences are empirical rather than derivable from the system's design by construction. The reduction in prompts per session is a measured behavior, not a fitted prediction. The only potentially relevant self-citation is [37], used in Section 3.1 to motivate the persona design ("This is inspired by an insight from [37]"), but it is not the evidence for the evaluation and does not constrain the outcome. The fact that the Orchid condition differs from the baseline on multiple design dimensions is a study-validity confound, not circularity. The paper makes no appeal to a uniqueness theorem or to a fitted input renamed as a prediction, and no derivation reduces to its own assumptions. The evaluation is self-contained against external expert judgments, so the circularity score is 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

No free parameters or invented entities: the paper reports a system evaluation, not a theoretical derivation. The assumptions above are about the validity and generalizability of the empirical measures.

assumptions (3)
  • domain assumption Expert ratings on 1-5 scales for novelty, feasibility, and value are a valid and reliable measure of creative outcome quality.
    Invoked in Section 4.5.1 (Measures) without reporting inter-rater reliability or validation against other creativity measures.
  • domain assumption The simulated work task scenario (Section 4.1) elicits creative behavior representative of real creative workflows.
    The study uses a 45-minute simulation with a provided market research report; generalization to longer, real projects is assumed.
  • domain assumption The baseline stack (web search, LLM chat, digital notebook) represents current practice for creative workflows.
    The baseline is based on a formative study, but the comparison is not an ablation; differences could come from workspace integration rather than context orchestration.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Orchid: Orchestrating Context Across Creative Workflows with Generative AI." pith.science (2026). https://pith.science/paper/GFEPFRPQ

@misc{pith2026250819517,
  author       = {Pith},
  title        = {Pith review of: Orchid: Orchestrating Context Across Creative Workflows with Generative AI},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GFEPFRPQ}},
  note         = {Machine review of arXiv:2508.19517}
}
read the original abstract

Context is critical for meaningful interactions between people and Generative AI (GenAI). Yet mainstream tools offer limited means to orchestrate it, particularly across workflows that span multiple interactions, sessions, and models, as often occurs in creative projects. Re specifying prior details, juggling diverse artifacts, and dealing with context drift overwhelm users, obscure intent, and curtail creativity. To address these challenges, we present Orchid, a system that gives its users affordances to specify, reference, and monitor context throughout evolving workflows. Specifically, Orchid enables users to (1) specify context related to the project, themselves, and different styles, (2) reference these via explicit mentions, inline selection, or implicit grounding, and (3) monitor context assigned to different interactions across the workflow. In a within-subjects study (n=12), participants using Orchid to execute creative tasks (compared to a baseline toolkit of web search, LLM-based chat, and digital notebooks) produced more novel and feasible outcomes, reporting greater alignment between their intent and the AI's responses, higher perceived control, and increased transparency. By prioritizing context orchestration, Orchid offers an actionable step toward next generation GenAI tools that support complex, iterative workflows - enabling creators and AI to stay aligned and augment their creative potential.

Figures

Figures reproduced from arXiv: 2508.19517 by the authors.

Figure 1
Figure 1. Orchid’s UI. a) Documents Sidebar. b) Editor Area. c) a result block from a Summary operation referencing context in a market research document using a persona’s style. Context is critical for meaningful interactions between people and Generative AI (GenAI). Yet mainstream tools o!er limited means to orchestrate it, particularly across work"ows that span multiple interactions, sessions, and models, as often occurs i… view at source ↗
Figure 2
Figure 2. "/" (Forward Slash) Commands: Users can access various creative operations by typing "/" to open the drop-down menu and [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Me page: Users can edit this text file to define how they would like to be seen by the system externalizing personal context [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: Users can provide a high-level objective, which the system can then break down into actionable tasks. Also, all following [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Example of a persona page. Users can edit this text file to customize their personas to best help with their work. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: This Summarize component is an example of a "/" operation. Here you can see how the user can make use of ’@’ mentions to [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: In-line prompting. This figure illustrates [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Transparency lens: Hovering over the generate bu [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: Critique component results. The figure illustrates how the result of a ”/” operation is presented. The [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: Orchid’s components. This figure illustrates at a high level how the front end and the back end interact. 1) when an operation takes place, the prompt visible in the component is sent to the back-end. 2) the back-end enhances the prompt into a meta-prompt using its kn…
Figure 11
Figure 11. Figure 11: Participants generated significantly more creative outcomes when using [PITH_FULL_IMAGE:figures/full_fig_p014_11.png]
Figure 12
Figure 12. Figure 12: Participants most frequently used explicit grounding via “@” mentions (M = 24.75), followed by implicit grounding in prompt [PITH_FULL_IMAGE:figures/full_fig_p015_12.png]
Figure 13
Figure 13. Figure 13: Participants reported stronger alignment and continuity when using Orchid compared to the baseline. Ratings show a higher [PITH_FULL_IMAGE:figures/full_fig_p016_13.png]
Figure 14
Figure 14. Figure 14: Participants frequently edited simulated personas, with moderate use of the Me page, and persona generation, suggesting [PITH_FULL_IMAGE:figures/full_fig_p017_14.png]
Figure 15
Figure 15. Figure 15: A greater proportion of participants agreed that Orchid helped align outputs with their style preferences and provided [PITH_FULL_IMAGE:figures/full_fig_p017_15.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 24 canonical work pages

  1. [1]

    Teresa M Amabile. 1996. Creativity and innovation in organizations . Vol. 5. Harvard Business School Boston

    work page 1996

  2. [2]

    Tyler Angert, Miroslav Suzara, Jenny Han, Christopher Pondoc, and Hariharan Subramonyam. 2023. Spellburst: A Node-based Interface for Exploratory Creative Coding with Natural Language Prompts. In Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology (San Francisco, CA, USA) (UIST ’23). Association for Computing Machinery, ...

    work page 2023

  3. [3]

    Peters, and Arman Cohan

    Iz Beltagy, Matthew E. Peters, and Arman Cohan. 2020. Longformer: The Long-Document Transformer. arXiv preprint arXiv:2004.05150 (2020). arXiv:2004.05150 [cs.CL]

    arXiv 2020

  4. [4]

    Pia Borlund. 2003. The IIR evaluation model: a framework for evaluation of interactive information retrieval systems. Information research 8, 3 (2003), 8–3

    work page 2003

  5. [5]

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners. Advances in neural information processing systems 33 (2020), 1877–1901

    2020

  6. [6]

    Aydar Bulatov, Yury Kuratov, and Mikhail Burtsev. 2022. Recurrent memory transformer. Advances in Neural Information Processing Systems 35 (2022), 11079–11091

    work page 2022

  7. [7]

    Kathy Charmaz. 2014. Constructing Grounded Theory (2nd ed.). Sage, London

    work page 2014

  8. [8]

    Herbert H Clark and Susan E Brennan. 1991. Grounding in communication. (1991)

    1991

Show all 47 references
  1. [9]

    Oliver Cobb and Arnaud Van Looveren. 2022. Context-Aware Drift Detection. arXiv: 2203.08644 [stat.ML] https://arxiv.org/abs/2203.08644

    2022 arXiv

  2. [10]

    British Design Council. [n.d.]. The Double Diamond. https://www.designcouncil.org.uk/our-resources/the-double-diamond/

  3. [11]

    Le, and Ruslan Salakhutdinov

    Zihang Dai, Zhilin Yang, Yiming Yang, Jaime Carbonell, Quoc V. Le, and Ruslan Salakhutdinov. 2019. Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context. arXiv: 1901.02860 [cs.LG] https://arxiv.org/abs/1901.02860

    2019 arXiv

  4. [12]

    K DEYA. 2001. Understanding and using context, personal and ubiquitous computing. Springer-Verlag 5, 1 (2001), 4–7

    2001

  5. [13]

    Gerhard Fischer. 2001. User modeling in human–computer interaction. User modeling and user-adapted interaction 11 (2001), 65–86

    2001

  6. [14]

    Jonas Frich, Jens Emil Grønbæk, Marcel Borowski, and Peter Dalsgaard. 2024. Exploring the impact of AI features on collaborative creativity. (2024)

    2024

  7. [15]

    Jonas Frich, Lindsay MacDonald Vermeulen, Christian Remy, Michael Mose Biskjaer, and Peter Dalsgaard. 2019. Mapping the landscape of creativity support tools in HCI. In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems . 1–18

    2019

  8. [16]

    Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yi Dai, Jiawei Sun, Qianyu Guo, Meng Wang, and Haofen Wang. 2023. Retrieval- Augmented Generation for Large Language Models: A Survey. ArXiv abs/2312.10997 (2023). https://api.semanticscholar.org/CorpusID:266359151

    2023 arXiv

  9. [17]

    Barbara J Grosz and Candace L Sidner. 1986. Attention, intentions, and the structure of discourse. Computational linguistics 12, 3 (1986), 175–204

    1986

  10. [18]

    Jonathan Grudin. 1994. Groupware and social dynamics: Eight challenges for developers. Commun. ACM 37, 1 (1994), 92–105

    1994

  11. [19]

    Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, and Ming-Wei Chang. 2020. REALM: Retrieval-Augmented Language Model Pre-Training. arXiv:2002.08909 [cs.CL] https://arxiv.org/abs/2002.08909

    2020 arXiv

  12. [20]

    Rossi, Subhabrata Mukherjee, Xianfeng Tang, Qi He, Zhigang Hua, Bo Long, Tong Zhao, Neil Shah, Amin Javari, Yinglong Xia, and Jiliang Tang

    Haoyu Han, Yu Wang, Harry Shomer, Kai Guo, Jiayuan Ding, Yongjia Lei, Mahantesh Halappanavar, Ryan A. Rossi, Subhabrata Mukherjee, Xianfeng Tang, Qi He, Zhigang Hua, Bo Long, Tong Zhao, Neil Shah, Amin Javari, Yinglong Xia, and Jiliang Tang. 2025. Retrieval-Augmented Generatio...

    2025 arXiv

  13. [21]

    World Leaders in Research-Based User Experience. [n.d.]. Thinking aloud: The Number 1 usability tool. https://www.nngroup.com/articles/thinking- aloud-the-1-usability-tool/

  14. [22]

    Dan Jurafsky and James H Martin. 2021. Speech and Language Processing . Prentice Hall. 3rd Edition Draft, available online. Orchid: Orchestrating Context Across Creative Work"ows with Generative AI 21

    2021

  15. [23]

    Jikun Kang, Wenqi Wu, Filippos Christianos, Alex J Chan, Fraser Greenlee, George Thomas, Marvin Purtorab, and Andy Toulis. 2025. LM2: Large Memory Models. arXiv preprint arXiv:2502.06049 (2025)

    2025 arXiv

  16. [24]

    Dae Hyun Kim, Hyungyu Shin, Shakhnozakhon Yadgarova, Jinho Son, Hariharan Subramonyam, and Juho Kim. 2024. AINeedsPlanner: A Workbook to Support E!ective Collaboration Between AI Experts and Clients. In Proceedings of the 2024 ACM Designing Interactive Systems Conference (Cope...

    2024

  17. [25]

    Avraham N Kluger and Angelo DeNisi. 1996. The e!ects of feedback interventions on performance: a historical review, a meta-analysis, and a preliminary feedback intervention theory. Psychological bulletin 119, 2 (1996), 254

    1996

  18. [26]

    Aobo Kong, Shiwan Zhao, Hao Chen, Qicheng Li, Yong Qin, Ruiqi Sun, Xin Zhou, Enzhi Wang, and Xiaohang Dong. 2024. Better Zero-Shot Reasoning with Role-Play Prompting. arXiv: 2308.07702 [cs.CL] https://arxiv.org/abs/2308.07702

    2024 arXiv

  19. [27]

    thinking-aloud

    Clayton Lewis. 1982. Using the" thinking-aloud" method in cognitive interface design . IBM TJ Watson Research Center Yorktown Heights, NY

    1982

  20. [28]

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in neural information processing s...

    2020

  21. [29]

    What it wants me to say

    Michael Xieyang Liu, Advait Sarkar, Carina Negreanu, Benjamin Zorn, Jack Williams, Neil Toronto, and Andrew D Gordon. 2023. “What it wants me to say”: Bridging the abstraction gap between end-user programmers and code-generating large language models. In Proceedings of the 202...

    2023

  22. [30]

    Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang

    Nelson F. Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. 2023. Lost in the Middle: How Language Models Use Long Contexts. arXiv: 2307.03172 [cs.CL] https://arxiv.org/abs/2307.03172

    2023 arXiv

  23. [31]

    Yiren Liu, Pranav Sharma, Mehul Jitendra Oswal, Haijun Xia, and Yun Huang. 2024. PersonaFlow: Boosting Research Ideation with LLM-Simulated Expert Personas. arXiv: 2409.12538 [cs.HC] https://arxiv.org/abs/2409.12538

    2024 arXiv

  24. [32]

    Maryellen C MacDonald, Neal J Pearlmutter, and Mark S Seidenberg. 1994. The lexical nature of syntactic ambiguity resolution. Psychological review 101, 4 (1994), 676

    1994

  25. [33]

    Narges Mahyar and Melanie Tory. 2014. Supporting communication and coordination in collaborative sensemaking. IEEE transactions on visualization and computer graphics 20, 12 (2014), 1633–1642

    2014

  26. [34]

    Gary Marcus. 2022. Stop Treating AI Models Like People . https://garymarcus.substack.com/p/stop-treating-ai-models-like-people?utm_source= substack&utm_medium=email Substack Newsletter

    2022

  27. [35]

    OpenAI. 2023. GPT-4 Technical Report. ArXiv abs/2303.08774 (2023). https://api.semanticscholar.org/CorpusID:257532815

    2023 arXiv

  28. [36]

    Patil, Ion Stoica, and Joseph E

    Charles Packer, Sarah Wooders, Kevin Lin, Vivian Fang, Shishir G. Patil, Ion Stoica, and Joseph E. Gonzalez. 2024. MemGPT: Towards LLMs as Operating Systems. arXiv: 2310.08560 [cs.AI] https://arxiv.org/abs/2310.08560

    2024 arXiv

  29. [37]

    Srishti Palani and Gonzalo Ramos. 2024. Evolving roles and work"ows of creative practitioners in the age of generative AI. In Proceedings of the 16th Conference on Creativity & Cognition . 170–184

    2024

  30. [38]

    Jun Rekimoto, Yuji Ayatsuka, and Kazuteru Hayashi. 1998. Augment-able reality: Situated communication through physical and digital spaces. In Digest of Papers. Second International Symposium on Wearable Computers (Cat. No. 98EX215) . IEEE, 68–75

    1998

  31. [39]

    Jeba Rezwana and Mary Lou Maher. 2023. User perspectives on ethical challenges in human-AI co-creativity: A design #ction study. In Proceedings of the 15th Conference on Creativity and Cognition . 62–74

    2023

  32. [40]

    Jeba Rezwana and Mary Lou Maher. 2024. Conceptual Models as a Basis for a Framework for Exploring Mental Models of Co-Creative AI. (2024)

    2024

  33. [41]

    Milad Saad and Mary Lou Maher. 1996. Shared understanding in computer-supported collaborative design. Computer-Aided Design 28, 3 (1996), 183–192

    1996

  34. [42]

    Joon Gi Shin, Janin Koch, Andrés Lucero, Peter Dalsgaard, and Wendy E Mackay. 2023. Integrating AI in human-human collaborative ideation. In Extended Abstracts of the 2023 CHI Conference on Human Factors in Computing Systems . 1–5

    2023

  35. [43]

    Robert J Sternberg. 1999. Handbook of creativity . Cambridge University Press

    1999

  36. [44]

    Lev Tankelevitch, Viktor Kewenig, Auste Simkute, Ava Elizabeth Scott, Advait Sarkar, Abigail Sellen, and Sean Rintel. 2023. The Metacognitive Demands and Opportunities of Generative AI. arXiv preprint arXiv:2312.10893 (2023)

    2023 arXiv

  37. [45]

    Michael Terry, Chinmay Kulkarni, Martin Wattenberg, Lucas Dixon, and Meredith Ringel Morris. 2023. AI alignment in the design of interactive AI: Speci#cation alignment, process alignment, and evaluation support. arXiv preprint arXiv:2311.00710 (2023)

    2023 arXiv

  38. [46]

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al . 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems 35 (2022), 24824–24837

    2022

  39. [47]

    Shunyu Yao, Je!rey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. ReAct: Synergizing Reasoning and Acting in Language Models. arXiv: 2210.03629 [cs.CL] Orchid: Orchestrating Context Across Creative Work"ows with Generative AI a A SUPPLEMENTARY MA...

    2023 arXiv

Pith tools

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