A Grey Literature Review of Graduate Students' Research Practices and Skills in the LLM Era

summary

The gist

Generative AI tools have rapidly learned in the daily workflow of graduate students in Software Engineering…

In short

The study analyzed student posts to find how graduate students use AI in research. It found students often outsource cognitive effort, leading to poor results and a lack of necessary skills like critical thinking and verification. Education needs to teach students how to effectively use LLMs responsibly.

Key concepts

False Content and Lack of Validation
This practice involves using AI to create fake references or generate text without checking if it is accurate. Students also use AI for text masking, which means hiding information in a way that might be misleading, showing a lack of critical validation in their work.
Autonomous Learning Skill
This crucial skill means students must not rely on AI to do everything they need to learn. It requires them to actively engage with the material, such as rewriting code from scratch after reading it, ensuring they truly understand the concepts rather than just copying AI output.
Managing LLM Context Skill
This refers to the ability of students to effectively manage conversations and prompts with LLMs. It includes knowing how to construct precise prompts and actively detecting when the AI is 'hallucinating' or making up facts, which is essential for reliable research.
Programming Skills (Technical Foundation)
This goes beyond just knowing how to write code superficially. It means having a solid technical base so students can actually reproduce code and understand the underlying logic, ensuring they don't just rely on AI to solve every technical problem.

Terminology used across episodes

This episode discusses

The paper

A Grey Literature Review of Graduate Students' Research Practices and Skills in the LLM Era · Read on arXiv

Enne Rebeca Silva de Freitas, Gustavo Pinto, Danilo Ribeiro

Universidade Federal do Pará · CESAR School

Context: Graduate students in computing now use Large Language Models (LLMs) throughout their research, from literature reviews and coding to data analysis and writing. Yet there is little shared understanding of which skills competent use of these tools requires, so graduate programs struggle to teach, mentor, or assess it, and unverified outputs can propagate into the scientific record. Objective: We investigate which LLM-based practices graduate students employ in their research and which skills they need to benefit from these practices. Method: We conducted a grey literature review of Reddit discussions. We retrieved 1,383 posts from five research-oriented subreddits, screened them against explicit inclusion and exclusion criteria, and analyzed the 54 included posts with thematic analysis, in which two authors independently coded each excerpt. Results: We identified seven categories of practices and nine categories of skills. Commenters locate the problem in accepting generated output without inspection rather than in using LLMs, and warn that delegating the effortful parts of a task, such as debugging, reading, or validating sources, delivers the output without building the skill the task was meant to develop. Traditional skills, such as programming and critical thinking, are what allow students to verify what an LLM produces. Conclusions: Our categories give graduate programs a concrete basis to teach students to work with LLMs without eroding the skills that research training should build, and they add research-specific skills, such as data sharing and scientific reproducibility, to existing AI competency frameworks.

Transcript

Introduction to the show: ident: Computer Science. The day in computer science: software, systems and theory on arXiv.

Tom: I'm Tom.

Jane: And I'm Jane, here with Lu, Meng and our in-house large language model, Lalam.

Tom: We're looking at the paper: "A Grey Literature Review of Graduate Students' Research Practices and Skills in the LLM Era".

Title and authors: Tom: So we’re looking at this paper today called "A Grey Literature Review of Graduate Students' Research Practices and Skills in the LLM Era." It sounds like it’s trying to figure out what actually matters when students are using AI for their research.

Jane: Yeah, it dives into the specific struggles graduate students have when they use these tools, not just general stuff about AI literacy. It’s important because a lot of the advice out there isn't tailored to this exact situation.

Lu: It starts by looking at what students are actually doing in their daily research workflow—things like writing, coding, and literature reviews—and how they use the AI tools for those tasks.

Meng: So it’s mapping out the actual practices so we can see where the problems lie when you try to teach these things properly.

Lalam: It basically analyzes a big set of posts from research-focused subreddits to see what students are doing with generative AI in their work, and then identifies what skills they are missing.

The paper's summary: Tom: So the core finding is that graduate students systematically outsource the hard cognitive work needed to develop actual research skills when they use these LLMs.

Jane: That’s a big deal because it means they end up without the expected results or with the necessary competence for real research. It points out that this outsourcing happens across almost every stage of their work.

Lu: The paper found that students are using AI for things like fabricating references, which is a form of false content, and they also use it to get help writing code or generating documentation.

Meng: And not just writing code; they’re using it for automated testing and quality assurance to explain how the code is structured. It’s not just one thing; it’s everywhere.

Lalam: So the paper lays out two main things: first, what practices students are actually employing with the LLMs, and second, what skills they need to be effective in this new research environment.

The paper's improvements: Tom: The authors point out that there’s a real gap between the current top-down lists of competencies and what graduate students themselves feel are missing.

Jane: They suggest that instead of just giving students a list, we need to use evidence grounded in what the students themselves perceive as the missing pieces. That makes it more practical for teaching.

Lu: They identify two major categories of skills needed: first, LLM-based practices, and second, the actual LLM-based skills themselves.

Meng: For those skills, they highlight something called the "autonomous learning skill," which means students have to stop outsourcing everything they need to learn from the AI.

Lalam: They also stress that traditional programming skills aren't enough; students need strong technical foundations beyond just being able to read code superficially.

Conclusion: Tom: So, the big idea is that graduate programs shouldn't take a binary choice between totally banning AI or letting them use it completely unrestricted.

Jane: Instead, they should focus on combining those traditional skills, like critical thinking and programming, with specific LLM skills like verifying the output you get back.

Lu: The central challenge for education is defining what responsible use looks like in research training—basically teaching students how to explain what the AI generates and how to verify those claims.

Meng: If we look at it from a production standpoint, the focus needs to be on ensuring that when a student uses AI for testing or documentation, they are actually verifying the routines themselves, not just accepting the output blindly.

Lalam: To wrap up this grey literature review of graduate students' research practices and skills in the LLM era, it shows that while productivity goes up with these tools, indiscriminate use without critical thinking leads to cognitive outsourcing.

More episodes

← Home