Vibe Scholarship

Vibe Scholarship: Accessible Academia or Intellectual Fraud?

Rated Collin’s dictionary word of the year, one key phrase of the informatic Zeitgeist of 2025 is ‘vibe coding’ (meske_vibe_2025?). Pilloried, occasionally, as part of a wider shift towards a mediatic world of ‘AI slop’, the term registers one of the effects of artificial intelligence: the erasure of an epistemic boundary that had previously cordoned off a complex formal activity for those possessing a particular expertise. Such expertise is usually acquired via channels of professional induction: the university, the occupational apprenticeship, or the institutional certificate. Software engineering has historically been one such field, although as terms like ‘hacking’ and ‘coding’ suggest, it is a professional activity marked by relative porous borders demarcating its own formality.

Meske et al. formalize ‘vibe coding’ as:

Vibe coding is a software development paradigm where humans and generative AI engage in collaborative flow to co-create software artifacts through natural language dialogue, shifting the mediation of developer intent from deterministic instruction to probabilistic inference. (Meske2025Vibe?)

[More on vibe coding…]

Semiotic Vibrations: From Coding to Scholarship

It is perhaps inevitable to ask how ‘vibes’ could translate from the world of coding to that of scholarship. Consider how much academic work, in the application of both quantitative and qualitative methods, involve forms of coding. Statistics invovles coding Python; thematic analysis involves coding texts. In the former case, academic coding always preempted, in its slack adherence to industry practices of documentation and testing for example, the chill and cavalier pose that Karpathy conjures up with the idea of ‘vibe coding’. In the latter, even as Braun and Clark’s famous discussion of coding became itself regimented – co-opted into epistemic ‘codes’ of rigor and replicability, to make for its peceived subjectivity relative to statistics – the idea of ‘vibes’ always lay in the background, part of the coder’s intuition, a hermeneutic impulse that could never be entirely subdued by the mechanized interfaces of systems that go by the appropriately technocratic acronym of “CAQDAS” (Computer-Aided Qualitative Data Analysis Systems).

Consider also the attraction of the term “resonance” in the humanity desciplines. Less likely to convey sun-drenched beaches than “vibration”, and certainly less than “vibe”, resonance endures as a sonic metaphor for the transmission of meaning. If something resonates, it hits home. A signifier arrives at its destination, even if that destination is hazy or fuzzy, a case of probabilistic approximation that also forecasts the stochastic nature of Large Language Models, whose success is based on whether its outputs resonate – have the right “feel” – as much as benchmark results.

Such clues point to the possibility of a concept of “vibe scholarship”. The obvious - and extreme - example of vibe scholarship is the plagiarised essay. If a LLM can “one-shot” a web application, it can of course also produce a research paper through the summons of a prompt. In the case of coding, where the question of authorial attribution and “co-creation” is not always at issue, plagiaristic “vibe coding” represents less of a disruption. Conversely in the world of scholarship, plagiarism strikes at the heart at its enterprise. This example marks a limit at which AI use cannot be santioned. But are there other uses?

At the other extreme, AI serves as a minimal advance on existing digital tools for checking spelling and grammar. University policies permit these uses, though preferably with acknowledgement. But vibe scholarship as it is envisioned here also considers a concerning middle ground: a set of uses that operate within the fuzziness of acceptability of the process of ‘co-creation’ (Meske2025Vibe?) describe above. These include many well-documented AI operations, more or less also operating within most policy limits: summarizing and explaining other articles, structuring and reviewing one’s own article.

These are standard AI uses, with nothing necessarily ‘vibration’-related. But there are other types of AI use that lean into the metaphor more directly. Some examples include:

• Asking a LLM to review a set of PDF articles and design a Mermaid diagram that renders their relationships
• Develop a cartographic map of a scholar’s career, charting the turns and changes by publication
• Generate a set of contrastive headings for a literature review, based on chronological, thematic or developmental organization of that literature
• Develop and apply a set of codes based on a corpus of either academic literature or primary research documents – interview transcripts, ethnographic field notes, and so on

Summarzing such uses, if Meske et al.’s formulation were to be translated to the world of scholarship, it might read as:

Vibe scholarship is a academic paradigm where humans and generative AI engage in collaborative flow to co-create research artifacts through natural language dialogue, shifting the mediation of developer intent from deterministic instruction to probabilistic inference. (Meske2025Vibe?)

From Coding to Scholarship

What about the domain of scholarship? Can artificial intelligence – in particular, large language models (LLMs) such as OpenAI’s ChatGPT, Anthropic’s Claude or DeepSeek – be applied to similar purposes in a field that has sought to maintain, via practices of peer review, citation curation, and journal quality management, much tighter control over who and what is permitted through its gates? Conversely – since it is already clear that LLM use is rife within academia – can that use be legitimate and if so how?

Borrowing from Karpathy (Karpathy2025How?), this article introduces the idea of ‘vibe scholarship’. What is this idea, and how can it be distinguished from related terms like ‘vibe research’ (Bertina2025VibeResearch?)?

Following Meske’s formulation above, we can start with the following definition:

Vibe scholarship is an approach to academic work where humans and generative AI collaborate to co-create knowledge through natural language dialogue, with emphasis on that collaboration being gestural, simulated, and granular.

This definition modifies several properties of Meske et al’s. Critically, the question of mediation of intent is dropped - or perhaps, deferred – as the distinction between artefacts (software or otherwise) and knowledge in part involves the very question of determination.

Equally important, three properties are added, designed to demarcate vibe scholarship from plagiarism. The first involves the idea of LLM support for scholarship at a gestural level: as a tool for indicating, mapping out, sketching, drafting, thinking aloud, without committing to one or another specific path. The second involves LLM use as a tool for simulation: for practicing, rehearsing, anticipating, enacting research in a way that thinks about some of its possibilities and pitfalls before these eventuate. The third invoves LLM use at a fine rather than coarse-grained level: most obviously, not using LLMs to write whole arguments – or indeed any part of an eventual contribution – but rather to work at the most applied level of scholarship. The more evident examples here might be pre-AI computer tool support for scholarship, in the form of checking spelling and grammar, formating citations, or producing graphs based on quantitative or qualitative data.

In the remainder of this article, this definition is used to describe how ‘vibe scholarship’ can be used to characterise legitimate and indeed generative use of AI in academic research, with particular emphasis on the types of qualitative and quantitative research carried out in the social sciences.

Lifecycle of Research

Vibe scholarship is discussed here in relation to what is often treated, in varied ways, as the lifecycle of social research in classic and widely cited methodology texts (booth_craft_2009?, bryman, yin, braun_and_clark). Though approaches naturally vary, common features of social research include

Conclusion

References