Sid’s Literature Review
[2512.22418] Building Software by Rolling the Dice: A Qualitative Study of Vibe Coding
Building Software by Rolling the Dice: A Qualitative Study of Vibe Coding: This paper examines the rise in LLM-assisted code generation amongst developers through prompting rather than actual code. Zhang talks about the recent phenomenon “vibe coding,” where programmers basically trust AI-generated code rather than deep understanding of the codebase and underlying logic. The study used qualitative interviews and observational analysis of developers who use AI coding tools to understand workflows, make critical code decisions, debug systems, etc. At a high level, the data depicted many instances where users accept incorrect or suboptimal code because it “feels right” and works superficially – this drastically increases what we call ‘tech debt’ where there may be thousands of lines of code that serve no practical purpose or actively even determinant the system but remain in the codebase due to a lack of understanding and fear of breaking what currently works. The authors find that this over-reliance reduces debugging rigor and long-term code quality as going from 0-to-1 becomes easy but scaling the system becomes notoriously difficult. This tells us that while AI increases development speed initially, it erodes actual software skills and increases tech debt to dangerous levels.
The effects of over-reliance on AI dialogue systems on students’ cognitive abilities: a systematic review | Smart Learning Environments | Springer Nature Link
The effects of over-reliance on AI dialogue systems on students’ cognitive abilities: a systematic review: This review research article on AI learning tools like ChatGPT reveal many similar themes to the article above. The authors share that frequent AI use negatively affects higher-order cognitive skills such as critical thinking and problem-solving. The paper, which basically follows a literature review methodology, analyzed studies across education levels/disciplines; the data actually showed mixed short-term performance gains but unanimous drops in independent reasoning and cognition. The researchers deduced that over-reliance on said AI tools inevitably leads to shallow learning, which leads to reduced retention, and ultimately leads to reduced cognitive ability. This means that without intentional design for the sake of enhancing human abilities, an over-reliance on AI runs a dangerous risk.
Eva’s Literature Review
How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use
The paper by Cathy Mengying Fang, Auren R. Liu, and others on How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use: A Longitudinal Randomized Controlled Study conducted a four-week randomized controlled trial with 981 participants, who generated over 300,000 messages to investigate how different AI chatbot interaction modes and conversation types affect mental well-being. The researchers tested three interaction modes: text-only, neutral voice, engaging voice, and three conversation types: open-ended, non-personal tasks, personal topics, in a 3×3 factorial design, measuring loneliness, real-world socialization, emotional dependence on AI, and problematic AI usage. The experimental conditions showed minimal effects on psychosocial outcomes, but the paper revealed that participants who voluntarily spent more time with chatbots, regardless of assigned condition, consistently showed worse outcomes across all measures, and individual characteristics like higher trust in AI and perceiving it as a friend were associated with increased emotional dependence and problematic use. This is important for us because it highlights one of our insights from our interviews that cognitive decline increases with prolonged usage of AI.
U.S. High School Students’ Use of Generative Artificial Intelligence
The paper by Alexandra Adair, Jessica Howell, Amanda Jacklin, and Alexandria Walton Radford in the U.S. High School Students’ Use of Generative Artificial Intelligence examines Generative AI usage among U.S. high school students, parents, and educators through surveys conducted from June 2024 to June 2025. The paper found that student Generative AI use for schoolwork increased from 79% to 84% between January and May 2025, with ChatGPT being the dominant platform, having a 69% usage rate, and primary applications include brainstorming, research, and essay editing. Moreover, students, parents, and educators recognize benefits of using Generative AI, with 57% of parents and over 85% of administrators viewing Generative AI use as valuable, but significant concerns persist, particularly around academic integrity due to cheating, and overreliance on technology weakening essential learning skills, and a worry about teacher preparedness for AI integration. For us, this means that we are solving an ongoing problem that is a reflection of the current education system’s ongoing struggle to establish coherent approaches to GenAI integration.
Natalia’s Literature Review
Exploring the effects of artificial intelligence on student and academic well-being in higher education: a mini-review – PMC
This article describes how AI in higher education has an impact on student well-being in beneficial and detrimental ways. Although there has been limited empirical evidence available, AI tools have been shown so far to support student academic success through personalized-learning, increased efficiency, better accessibility for learners with diverse needs, and mental health support through chatbots and virtual assistants. The benefits are associated with reduced academic stress, higher engagement, and better communication in the classroom. There are also findings that suggest that excessive reliance on AI could be harmful by increasing digital fatigue, loneliness, and decreasing face-to-face interactions. This could lead to social impairments and worse interpersonal skills and emotional intelligence. Some other concerns are related to data privacy, surveillance, and job replacement, which exacerbates anxiety among students and teachers. Overall, the studies reviewed in this article underscore the importance of ethical and human-centered AI integration in higher education. There is a call to maximize AI’s supportive capabilities while mitigating its unintended consequences for student’s mental, emotional, and social well-being.
The Usage of AI in Teaching and Students’ Creativity: The Mediating Role of Learning Engagement and the Moderating Role of AI Literacy – PMC
This article finds that using AI in teaching has a significant positive impact on students’ creativity, directly and indirectly. Specifically, AI-supported teaching methods make lessons more engaging for students, which promotes higher levels of creativity. Interestingly, students’ perceptions of their teachers’ AI literacy plays a moderating role in this process. When teachers are perceived as highly AI-literate, the positive impact of AI use in engagement, and on creativity as a consequence, is strengthened. On the other hand, lower perceived AI-literacy weakened these effects. The findings demonstrate that AI does not foster creativity automatically but does so through increased student engagement and effective, competent integration by teachers. The results found in this article highlight the importance of pedagogical design and teacher AI literacy in realizing the creative potential of AI-assisted teaching.
Angela’s Literature Review
The effect of ChatGPT on students’ learning performance, learning perception, and higher-order thinking: insights from a meta-analysis
This paper by authors Jin Wang and Wenxiang Fan performs a meta-analysis of 51 research studies in order to assess the effectiveness of ChatGPT in improving students’ learning performance, perception, and higher-order thinking. These studies mostly involved high school and college students. They included only experimental studies that compared experimental groups with control groups and synthesized their results with a meta-analysis software. They found that ChatGPT substantially improves students’ learning performance and moderately improves students’ learning perceptions and higher-ordering thinking. However, Chat had the most positive impact on learning performance only when the study lasted 4-8 weeks; Chat also had the weakest positive impact in project-based learning. Furthermore, Chat’s ability to foster higher-order thinking varied across different types of courses, performing better in STEM classes. All in all, Chat functioned the best when it was an “intelligent tutor.” This study is important because it demonstrates the upsides of AI in learning – which means that our attempt to create “healthy” AI can actually be a positive thing.
Artificial intelligence in education: a systematic literature review
This paper by __ overviews the current state of research on artificial intelligence in education by applying a bibliometric analysis to 2,223 research articles and a content analysis to 125 papers. They find that the overall research landscape is guided by a few key concepts – for example, intelligent tutoring systems, constructivist learning theory, and intelligent assessment systems. They also reveal some deficits in the research: preschool education and ethical considerations. Very few AIED articles assess the ramifications of AIED’s ethical risks – they suggest incorporating not only performance metrics but ethical metrics like fairness, algorithm transparency, and trust. This study is important to us because it highlights the importance of our kind of research: research that tries to “harm reduce” AI.
Varsha’s Literature Review
Beware of Metacognitive Laziness: Effects of Generative Artificial Intelligence on Learning Motivation, Processes, and Performance
In this paper on Metacognitive Laziness as a result of Generative Artificial Intelligence, Fan et al. (2025) studied how GenAI affects learning motivation, self-regulated learning (SRL), and academic performance in an academic writing task. The researchers discuss the concept of hybrid intelligence and introduce “metacognitive laziness” , a phenomenon where learning overrely on AI and inadvertently reduces their own metacognitive effort. To test this phenomenon, this study collected experimental data from 117 university students that were randomly assigned to four groups (ChatGPT, human expert, checklist tools, or no support) and measured motivation, learning measures, essay scores, and performance on knowledge tests. The authors found that while ChatGPT considerably enhanced short term writing performance, it did not increase knowledge gain/transfer or intrinsic motivation. Further, learners in the AI group engaged less in metacognitive processes compared to those in other groups. Findings show that while generative AI can boost immediate task performance, it may undermine deeper learning and self regulation, which suggests that educators should design scaffold tools to prevent overreliance on AI.
Scaffold or Crutch? Examining College Students’ Use and Views of Generative AI Tools for STEM Education
Wang et l. (2024) studied how university students use generative AI tools in STEM education and what impact these tools have on problem solving practices. The researchers discuss how GenAI may transform STEM problem solving by acting either as a scaffold (which generally supports learning) or a crutch (which generally limits learning in students by bypassing cognitive effort). The study collected survey data from 40 STEM students and 28 STEM instructors, analyzing usage patterns, prompting behaviors, and their perceived risks and benefits of these tools. The authors found that students widely use AI tools for summarizations, explanations, solving problem sets, which instructors warned against misinformation, overreliance and violations of academic integrity. Findings indicate that although GenAI is deeply integrated into student workflows, its current use often bypasses the struggle involved in authentic problem solving processes. This calls to action the role of educators in building instruction and tool designs that encourage scaffolding rather than solution generation.
