Large Language Models in Architecture Studio: A Framework for Learning Outcomes

Juan David Salazar Rodriguez; Julfendi; Khoo Eng Tat; Nachamma Sockalingam; Sam Conrad Joyce

REVIEW 4 major objections 4 minor 16 references

Large Language Models in Architecture Studio: A Framework for Learning Outcomes

T0 review · 4 major / 4 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read A framework positions large language models as co-teachers in the architecture studio, aligning AI interventions with Bloom's taxonomy to expand design education beyond form generation.

desk verdict Solid conceptual synthesis of LLM roles in the architecture studio; the load-bearing transfer assumption about visual competence is unvalidated and should be tested before the framework is treated as evidence. read the letter →

arxiv 2510.15936 v3 pith:HBAS3KOR submitted 2025-10-08 cs.CY cs.HC

Juan David Salazar Rodriguez , Sam Conrad Joyce , Nachamma Sockalingam , Khoo Eng Tat , Julfendi This is my paper · ORCID

classification cs.CYcs.HC

keywords largelanguagemodelsarchitecturestudiopedagogicalframeworkBloom'staxonomydesigneducationAIco-teacherself-directedlearningpeercritique

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that AI in architectural education has been used mainly as a co-designer—generating forms, automating tasks, improving representation—while the studio's real pedagogical core, learning to think like a designer, has been neglected. It identifies concrete challenges in three modes of studio learning—self-directed, peer-to-peer, and professor-led—and proposes specific large language model (LLM) interventions for each. These interventions are then aligned with Bloom's taxonomy so that learning outcomes become measurable rather than aspirational. If the framework holds, LLMs become co-teachers that strengthen student autonomy, collaboration, and self-reflection, and the studio's success is measured by designer development, not just design output.

What carries the argument

The framework rests on three organizing structures. First, seven studio activities—open-ended projects, rapid iteration, formal and informal critique, heterogeneous issues, precedent study, faculty-set constraints, and diverse media—are grouped into the double-diamond design process. Second, these activities are classified into three learning modes: self-directed, peer-to-peer, and professor-led, each with its own failure modes. Third, Bloom's taxonomy (a six-level scale of cognitive skills: remember, understand, apply, analyze, evaluate, create) supplies the scale that turns proposed LLM interventions into measurable learning outcomes. The load-bearing move is the one-to-one mapping between

What would settle it

A controlled semester-long studio study comparing a cohort using the proposed LLM interventions against a cohort without them, measuring metacognitive awareness inventories, critique-quality rubrics, and peer-assessment impartiality; if the intervention cohort shows no gains on these instruments, the framework's central mapping fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that large language models can be repositioned from co-designers to co-teachers in the architecture studio, and that this repositioning can be systematized. Building on seven fundamental studio activities and three modes of learning, it maps well-documented pedagogical pain points—cognitive overload in self-study, bias and superficiality in peer critique, power imbalances and jargon in jury feedback—to specific LLM affordances such as adaptive questioning, bias-mitigated feedback, jargon translation, and jury simulation. Each mapping is tied to a level of Bloom's taxonomy, so that the proposed interventions carry explicit, assessable learning outcomes. The result

Load-bearing premise

The framework assumes that LLM affordances demonstrated in other educational settings will transfer to the architecture studio's specific conditions—open-ended projects, power-laden juries, and visual/precedent-heavy reasoning—without empirical evidence in studio contexts.

Editorial extensions

If this is right

LLM tools can provide personalized, just-in-time feedback and adaptive cognitive scaffolding for self-directed studio work, supporting precedent recall and critical analysis.
In peer learning, LLMs can act as neutral mediators that structure feedback, reduce social bias, and train argumentation before live juries.
In professor-led learning, LLMs can translate jargon, interpret ambiguous critique, simulate jury personas, and free instructors for more creative guidance.
These interventions can be aligned with Bloom's taxonomy, giving studios a way to measure learning outcomes beyond design production.
AI integration in the studio need not replace the instructor; it strengthens the teaching role and expands student autonomy.

Reading between the lines

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

If the framework is correct, architecture accreditation and assessment could shift toward evaluating metacognitive skills (self-regulation, reflection, critique literacy) alongside portfolio quality.
The same challenge-to-intervention-to-taxonomy mapping could transfer to other project-based studio disciplines—industrial design, urban planning, engineering capstones—where open-ended briefs and desk crits dominate.
A testable extension: instrument a studio with and without LLM critique mediators and measure whether peer-feedback impartiality improves using the rubric-based metrics the paper cites.
The framework implicitly predicts that LLM support will accelerate lower-order Bloom levels (recall, understand) so instructors can concentrate on higher-order synthesis and evaluation; measuring time-on-task per taxonomy level could confirm this.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Quick take: this is a genuinely useful synthesis for architecture educators. Kuhn's seven studio activities, Bloom's taxonomy, and LLM affordances are all established pieces, but the integrated mapping across self-, peer-, and professor-led learning is the real contribution. The comparative tables in Section 4 give instructors something concrete to try, and the paper is transparent that it is conceptual—it says so explicitly in Section 2 ('eminently conceptual in nature').

What it does well: it names documented challenges in studio pedagogy (peer feedback bias, autonomy struggles, critique anxiety) and maps them to specific LLM actions with Bloom levels attached. The discussion of Kuhn's activities as overlapping rather than a rigid sequence is a nice corrective. The reference list is broad and appropriate.

Soft spots, in order of size. First, the load-bearing transfer assumption: the paper asserts LLMs 'can' support precedent recall, visual-storytelling, and design validation, but the cited evidence comes from text-dominant contexts—essay feedback, math tutoring, software tutorials. The stress-test note has the right concern: LLMs remain weak at the visual and spatial reasoning central to architectural precedent study. The paper itself calls for 'real study contexts' but gives no pilot or worked example with actual studio artifacts. That leaves the framework as a set of plausible hypotheses, not tested interventions. Second, the abstract and conclusions use 'findings' and 'shows' when nothing empirical was collected; careful revision should replace those with 'proposes' or 'hypothesizes.' Third, the literature synthesis is described as systematic, but no search protocol or inclusion criteria are given. Minor for a conceptual paper, but it is a mismatch to call it systematic.

None of these are fatal. The framework is internally coherent, the Bloom mapping is interpretive but not forced, and the authors are honest about needing validation. The multimodal weakness matters most where the framework targets precedent and visual narrative; those claims should be narrowed or tested before being sold.

Who this is for: architecture and design educators, and AIED researchers working on studio-based learning. It deserves serious peer review—send it out—but with the expectation that the authors will temper the claims and ideally add a small feasibility study.

Referee Report

4 major / 4 minor

Summary. This conceptual paper proposes a framework for using large language models (LLMs) as pedagogical tools in the architectural design studio. It builds on Kuhn's (2001) seven studio activities, maps them onto self-, peer-, and professor-led learning modes, and aligns challenges and proposed LLM interventions with the six cognitive levels of Bloom's taxonomy. The paper identifies documented pedagogical challenges (e.g., low self-regulation, biased peer feedback, critique anxiety), proposes LLM-based actions for each learning mode (e.g., precedent recall prompts, virtual peer debates, jargon translation), and claims these can support measurable learning outcomes. The framework is explicitly labeled as conceptual, with a recommendation for future empirical testing in real studio contexts.

Significance. If validated, the framework would make a useful contribution by shifting the discourse on AI in architecture education from form generation to pedagogy, and by providing a structured mapping that could guide future empirical work and curriculum design. The paper's strengths include its explicit grounding in the studio-activity literature (Kuhn, Schön), the use of Bloom's taxonomy as an organizing structure, and a clear visual presentation (Figures 1–9). It also candidly acknowledges its conceptual nature and the need for empirical validation. However, the central claim that LLMs can effectively support architecture-studio learning rests largely on evidence from adjacent educational domains that are text-dominant; the architecture studio's visual, spatial, and precedent-heavy reasoning is not addressed with direct evidence. The framework is a reasonable hypothesis-generating scaffold, but its correctness and practical utility remain unsubstantiated.

major comments (4)

[§4.1, §4.2] The proposed LLM actions require visual and spatial competence that the cited evidence does not support. For example, §4.1 claims LLMs enable 'systematic precedent recall by converting textual and visual data into structured prompts' and 'visual-storytelling development,' and §4.2 proposes 'bias-mitigated feedback' in peer critiques. The cited sources are predominantly text-based educational applications (essay feedback, math tutoring, software tutorials), not architectural plan/section/model interpretation. LLMs are primarily text processors; multimodal abilities are recent and unvalidated for domain-specific architectural critique. Without evidence that LLMs can meaningfully process architectural images, spatial organizations, or precedent semantics, the proposed interventions reduce to generic textual coaching, which does not address the visual core of studio learning. This is a load-
[§2 Methodology] The methodology is described as a 'systematic and critical review,' but no protocol is reported: no databases' search strings beyond three example terms, no inclusion/exclusion criteria, no screening process, and no count of sources considered. The mapping of challenges to Bloom's taxonomy also appears interpretive, with no rubric or inter-coder validation. This undermines the reproducibility of the challenge list and the 'measurable learning outcomes' claim. The authors should either provide a transparent review protocol or explicitly label the synthesis as narrative/scoping, with corresponding limitations.
[§3.2] The paper acknowledges that it provides no specific challenges for the 'Remember' level in peer learning (§3.2, final paragraph), leaving a gap in the Bloom mapping for one of the three learning modes. Since the framework's contribution is the claim of alignment with all six Bloom levels, this omission should be either filled with evidence or explicitly discussed as a boundary condition. As written, the mapping for peer learning is incomplete, and the subsequent proposals in §4.2 do not clearly address this level.
[Abstract and §5/§7] The abstract states that 'LLMs are emerging as complementary agents capable of generating personalized feedback, organizing collaborative interactions, and offering adaptive cognitive scaffolding,' and conclusions (§7) assert 'LLMs are a promising tool for complementing teaching and learning processes.' These statements go beyond the evidence presented, since the paper itself (Section 2) recommends testing 'in real study contexts' and provides no empirical data. The claims should be hedged to 'could' or 'may' and explicitly framed as hypotheses to be tested, not as demonstrated capacities.

minor comments (4)

[Throughout] Several typos and formatting issues: 'representational-al efficiency' (Abstract), 'hypo-thetical' (Abstract), 'size up (does this mean evaluate?)' left as an editing query (§3), 'Afa-can' instead of 'Afacan' (§3.3).
[References] Some citations are missing or inconsistent: 'Kaithe et al., 2025' is cited in §4.1 but not in the reference list; 'Kimet et al., 2023' (§4.2) appears as 'Kim et al.' in references; Matzakos & Moundridou (2025) is a civil-engineering mathematics laboratory, not architecture studio. Also, several refereed journal articles are conflated with arXiv preprints; the paper should distinguish peer-reviewed evidence from preprints.
[§5 and §7] The paper refers to 'Chapter 5' (§1) but later uses 'Section 5' as Discussion; section numbering should be consistent. Figure 9 is labeled 'Main challenges and learning outcomes' but no actual table is visible in the text; the authors should ensure all figures/tables are presented and cited.
[Figure 2-5] The cognitive-challenge domain figures (Figures 3–5) are small and not described in the text; consider enlarging them or providing table summaries so readers can follow the mapping.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an explicitly conceptual literature-based framework with no fitted predictions and no load-bearing self-citations.

full rationale

The paper does not derive quantitative predictions from fitted parameters, so the classic circularity failure modes do not apply. Its central structure is a mapping: Kuhn's seven studio activities are categorized into self-, peer-, and professor-led learning modes, challenges are identified from the cited empirical and theoretical literature, and LLM-based interventions are proposed as hypotheses to address those challenges, aligned with Bloom's taxonomy. The interventions are attributed to external sources (e.g., Kumar et al. 2024; Guo 2024; Fan et al. 2024) rather than to the authors' own prior results, and no author self-citations appear in the reference list. The paper explicitly disclaims empirical validation, stating that 'the proposed framework is eminently conceptual in nature' and recommending that 'subsequent research test the identified AI tools in real study contexts' (Section 2, Alignment with AI tools Phase). Thus the Bloom alignment is an interpretive organizational step, not a quantity extracted from data, and the challenge-to-intervention pairing is a design proposal rather than a circular derivation. The main weakness—whether LLM affordances transfer from text-dominant contexts to the visual, precedent-heavy architecture studio—is a correctness or external-validity concern, not a circularity concern.

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

No fitted numerical parameters or invented entities. The paper's load-bearing assumptions are domain-level: Kuhn's activity model, Bloom's taxonomy, the representativeness of the literature search, and the transferability of LLM educational affordances to architecture studios. The last is explicitly acknowledged by the authors as awaiting empirical validation.

assumptions (4)

domain assumption Architecture studio pedagogy is adequately captured by Kuhn's seven fundamental activities.
Section 3 builds the entire challenge taxonomy on Kuhn (2001) without testing completeness or overlap against alternative studio models.
domain assumption Bloom's taxonomy is the appropriate scheme for connecting pedagogical interventions to measurable learning outcomes.
Sections 3–5 map every challenge and LLM action onto Bloom levels; no justification is given for choosing Bloom over other outcome frameworks.
domain assumption LLM affordances demonstrated in other educational contexts will transfer to architecture studio tasks.
Section 4 cites LLM feedback/reflection studies from writing, math, engineering, and generic AIED; the paper assumes these apply to precedent analysis, desk crits, and juries. The paper itself flags this as needing empirical testing (Section 2, Section 7).
domain assumption The literature retrieved with the stated search terms is representative of the studio challenge space.
Section 2 lists databases and search terms but no inclusion/exclusion protocol, so the resulting challenge list depends on unreported selection judgments.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Large Language Models in Architecture Studio: A Framework for Learning Outcomes." pith.science (2026). https://pith.science/paper/HBAS3KOR

@misc{pith2026251015936,
  author       = {Pith},
  title        = {Pith review of: Large Language Models in Architecture Studio: A Framework for Learning Outcomes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HBAS3KOR}},
  note         = {Machine review of arXiv:2510.15936}
}

read the original abstract

The study explores the role of large language models (LLMs) in the context of the architectural design studio, understood as the pedagogical core of architectural education. Traditionally, the studio has functioned as an experiential learning space where students tackle design problems through reflective practice, peer critique, and faculty guidance. However, the integration of artificial intelligence (AI) in this environment has been largely focused on form generation, automation, and representation-al efficiency, neglecting its potential as a pedagogical tool to strengthen student autonomy, collaboration, and self-reflection. The objectives of this research were: (1) to identify pedagogical challenges in self-directed, peer-to-peer, and teacher-guided learning processes in architecture studies; (2) to propose AI interventions, particularly through LLM, that contribute to overcoming these challenges; and (3) to align these interventions with measurable learning outcomes using Bloom's taxonomy. The findings show that the main challenges include managing student autonomy, tensions in peer feedback, and the difficulty of balancing the transmission of technical knowledge with the stimulation of creativity in teaching. In response to this, LLMs are emerging as complementary agents capable of generating personalized feedback, organizing collaborative interactions, and offering adaptive cognitive scaffolding. Furthermore, their implementation can be linked to the cognitive levels of Bloom's taxonomy: facilitating the recall and understanding of architectural concepts, supporting application and analysis through interactive case studies, and encouraging synthesis and evaluation through hypothetical design scenarios.

Figures

Figures reproduced from arXiv: 2510.15936 by the authors.

**Figure 1.** Phased Methodology Literature Synthesis Phase. A systematic and critical review was conducted of classic and contemporary sources related to study pedagogy (Kuhn, 2001; Schön, 1983), Bloom's taxonomy (Bloom, 2010), and the application of AI in education. To this end, academic databases such as Scopus, Web of Science, and Google Scholar were consulted, using search terms such as “learning in architecture studies,” “p… view at source ↗

**Figure 6.** Applications of LLMs for Self-learning in architecture studio 4.2 Peer-Learning Actions Enabled by LLMs in Architecture Studios LLMs support virtual peer simulation by orchestrating debate scenarios where students defend or critique design decisions in real time. These simulated exchanges improve argumentation, metacognitive reflection, and reduce anxiety before live juries (Liu et al., 2024; Kjærsdóttir & Tzortzopo… view at source ↗

**Figure 7.** Applications of LLMs for Peer-learning in architecture studio 4.3 Professor-led Actions Enabled by LLMs in Architecture Studios Within professor-led studio settings, Large Language Models (LLMs) serve as pedagogical co-facilitators that augment the instructor’s feedback capacity, promote clarity in communication, and enhance students’ reflective understanding. As depicted in [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

16 extracted references · 2 linked inside Pith

[1]

The study explores the role of large language models (LLMs) in the context of the architectural design studio, understood as the pedagogical core of architectural education

Large Language Models in Architecture Studio: A Framework for Learning Outcomes Juan David Salazar Rodriguez1, Sam Conrad Joyce1, Nachamma Sockalingam2, Khoo Eng Tat3, Julfendi4 1 META Design lab, Architecture and Sustainable Design Pillar, Singapore University of Technology and Design, Singapore, Singapore 2 Office of Strategic Planning, Singapore Univer...

1987
[2]

learning in architecture studies,

Phased Methodology Literature Synthesis Phase. A systematic and critical review was conducted of classic and contemporary sources related to study pedagogy (Kuhn, 2001; Schön, 1983), Bloom's taxonomy (Bloom, 2010), and the application of AI in education. To this end, academic databases such as Scopus, Web of Science, and Google Scholar were consulted, usi...

2001
[3]

When studying the precedents, students struggle to remember, analyze and relevant precedents

Cognitive Challenge Domain of Self-learning The cognitive challenges of architectural studio activities for self-learning happens in their study of precedents, oscillating between big picture and details, performing rapid iteration with different medias and framing the problem of their project. When studying the precedents, students struggle to remember, ...

2021
[4]

Remember

Cognitive Challenge Domain of Peer-learning Architecture students face significant barriers when creating unified visual presentations during peer learning due to varied media skills. Disparities in IT 9 proficiency hinder effective collaboration, especially in virtual environments where peer-to-peer support is limited (Wang, 2023). Differences in visual ...

2023
[5]

The complexity of design problems requires students to manage multiple knowledge types, often leading to cognitive overload

Domain of Cognitive Challenge for Professor-led Learning At the Remembering level, students struggle to recall rule-based constraints, such as building codes and budget limits, during professor-led assignments. The complexity of design problems requires students to manage multiple knowledge types, often leading to cognitive overload. According to Chan (19...

1990
[6]

LLMs Actions and Learning Outcomes With the challenges across multiple levels, LLMs can be used to reshape the architecture studio education by supporting self-learning, peer-learning and professor-led activities. 4.1 Self-Learning Actions Enabled by LLMs in Architecture Studios LLMs strengthen systematic precedent recall by converting textual and visual ...

2025
[7]

These simulated exchanges improve argumentation, metacognitive reflection, and reduce anxiety before live juries (Liu et al., 2024; Kjærsdóttir & Tzortzopoulos, 2024)

Applications of LLMs for Self-learning in architecture studio 4.2 Peer-Learning Actions Enabled by LLMs in Architecture Studios LLMs support virtual peer simulation by orchestrating debate scenarios where students defend or critique design decisions in real time. These simulated exchanges improve argumentation, metacognitive reflection, and reduce anxiety...

2024
[8]

As depicted in Figure 8, these models strengthen the entire feedback cycle—from critique delivery to evaluative reasoning—across four pedagogical domains

Applications of LLMs for Peer-learning in architecture studio 4.3 Professor-led Actions Enabled by LLMs in Architecture Studios Within professor-led studio settings, Large Language Models (LLMs) serve as pedagogical co-facilitators that augment the instructor’s feedback capacity, promote clarity in communication, and enhance students’ reflective understan...

2025

Show all 16 references

[9]

Applications of LLMs for Professor-led learning in architecture studio 5 Discussion The study on the use of large language models (LLMs) in architecture studies opens up a field of pedagogical reflection that allows us to link the potential of artificial intelligence (AI) with...

2002
[10]

Main challenges and learning outcomes 7 Conclusions The incorporation of large-scale linguistic models into architectural education does not replace human interaction or design experience, but it does provide a robust framework to support learning processes. Its true value lie...

2021
[177]

• Agkathidis, A., Song, Y., & Ioanna, S. (2024). AI-Assisted Design: Utilising artificial intelligence as a generative form-finding tool in architectural design studio teaching. • Al-Qemaqchi, N. T. (2025). Biases shown by juries during architecture design assessment sessions....

2024
[356]

https://doi.org/10.3390/buildings15030356 • Zhou, X., Li, R., Liang, P., Zhang, B., Shahin, M., Li, Z., & Yang, C. (2025). Using LLMs in Generating Design Rationale for Software Architecture Decisions. arXiv preprint arXiv:2504.20781. • Zimmerman, B. (2002). Becoming a Self-Re...

2025
[1287]

Y., & Chen, W

• Ali, F., Choy, D., Divaharan, S., Tay, H. Y., & Chen, W. (2023). Supporting self-directed learning and self-assessment using TeacherGAIA, a generative AI chatbot application: Learning approaches and prompt engineering. Learning: Research and Practice, 9(2), 135–147. https://...

2023
[1613]

Embedded and Hopeful: A Curriculum for Change

• Keogh, S. Embedded and Hopeful: A Curriculum for Change. • Keenahan, J., & McCrum, D. (2021). Developing interdisciplinary understanding and dialogue between Engineering and Architectural students: design and evaluation of a problem-based learning module. European Journal of...

2021
[2971]

& Seo, M

• Kim, S., Shin, J., Cho, Y., Jang, J., Longpre, S., Lee, H. & Seo, M. (2023, October). Prometheus: Inducing fine-grained evaluation capability in language models. In The Twelfth International Conference on Learning Representations. • Kuhn, S. (2001). Learning from the archite...

2023 arXiv
[3928]

• Boarin, P., Martinez-Molina, A., & Juan-Ferruses, I. (2020). Understanding students’ perception of sustainability in architecture education: A comparison among universities in three different continents. Journal of Cleaner Production, 248, 119237. • Bloom, B. S. (2010). A ta...

2020 arXiv

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.

[1] [1]

The study explores the role of large language models (LLMs) in the context of the architectural design studio, understood as the pedagogical core of architectural education

Large Language Models in Architecture Studio: A Framework for Learning Outcomes Juan David Salazar Rodriguez1, Sam Conrad Joyce1, Nachamma Sockalingam2, Khoo Eng Tat3, Julfendi4 1 META Design lab, Architecture and Sustainable Design Pillar, Singapore University of Technology and Design, Singapore, Singapore 2 Office of Strategic Planning, Singapore Univer...

1987

[2] [2]

learning in architecture studies,

Phased Methodology Literature Synthesis Phase. A systematic and critical review was conducted of classic and contemporary sources related to study pedagogy (Kuhn, 2001; Schön, 1983), Bloom's taxonomy (Bloom, 2010), and the application of AI in education. To this end, academic databases such as Scopus, Web of Science, and Google Scholar were consulted, usi...

2001

[3] [3]

When studying the precedents, students struggle to remember, analyze and relevant precedents

Cognitive Challenge Domain of Self-learning The cognitive challenges of architectural studio activities for self-learning happens in their study of precedents, oscillating between big picture and details, performing rapid iteration with different medias and framing the problem of their project. When studying the precedents, students struggle to remember, ...

2021

[4] [4]

Remember

Cognitive Challenge Domain of Peer-learning Architecture students face significant barriers when creating unified visual presentations during peer learning due to varied media skills. Disparities in IT 9 proficiency hinder effective collaboration, especially in virtual environments where peer-to-peer support is limited (Wang, 2023). Differences in visual ...

2023

[5] [5]

The complexity of design problems requires students to manage multiple knowledge types, often leading to cognitive overload

Domain of Cognitive Challenge for Professor-led Learning At the Remembering level, students struggle to recall rule-based constraints, such as building codes and budget limits, during professor-led assignments. The complexity of design problems requires students to manage multiple knowledge types, often leading to cognitive overload. According to Chan (19...

1990

[6] [6]

LLMs Actions and Learning Outcomes With the challenges across multiple levels, LLMs can be used to reshape the architecture studio education by supporting self-learning, peer-learning and professor-led activities. 4.1 Self-Learning Actions Enabled by LLMs in Architecture Studios LLMs strengthen systematic precedent recall by converting textual and visual ...

2025

[7] [7]

These simulated exchanges improve argumentation, metacognitive reflection, and reduce anxiety before live juries (Liu et al., 2024; Kjærsdóttir & Tzortzopoulos, 2024)

Applications of LLMs for Self-learning in architecture studio 4.2 Peer-Learning Actions Enabled by LLMs in Architecture Studios LLMs support virtual peer simulation by orchestrating debate scenarios where students defend or critique design decisions in real time. These simulated exchanges improve argumentation, metacognitive reflection, and reduce anxiety...

2024

[8] [8]

As depicted in Figure 8, these models strengthen the entire feedback cycle—from critique delivery to evaluative reasoning—across four pedagogical domains

Applications of LLMs for Peer-learning in architecture studio 4.3 Professor-led Actions Enabled by LLMs in Architecture Studios Within professor-led studio settings, Large Language Models (LLMs) serve as pedagogical co-facilitators that augment the instructor’s feedback capacity, promote clarity in communication, and enhance students’ reflective understan...

2025