Slopsquatting is the creation of web sites or software packages with names that have not already been registered but that large language models (LLMs) are likely to hallucinate as the sources and the names of code libraries that they refer to when they are used to write computer programs. It is a development of typosquatting, where the sites or packages are registered under misspellings of existing project names that might be caused by human carelessness – or by cosmic rays flipping bits in modern computer memory.
Squatting in this sense is an exploit, it is hacking. It exploits errors in code, however caused, in order to take control of a system or to otherwise compromise it. It does so before the fact, laying the groundwork to exploit errors that are likely to be made in future. This is in contrast to more traditional hacking techniques that explore existing systems for bugs to exploit after they are released.
As well as writing software, LLMs are increasingly used in academia to write essays – usually without this fact being revealed. Just as they hallucinate references to software that doesn’t exist, LLMs are now producing essays that cite other essays which do not actually exist. Citations are by definition incomplete but this seems to take that too far for non-Lacanian academics, who are concerned with what they see as an epistemic crisis caused by the use of LLMs.
There is an opportunity here.
We can determine which imaginary essays an LLM is likely to hallucinate citations of by repeatedly prompting it to write essays on common topics. We can then write and publish “preprints” of those essays. Readers of LLM-generated essays will then search for these essay titles, and will find the ones that we have written.
What we do with these essays is up to us. We can simply improve our research ranking. We can close the holes in the episteme that LLMs would otherwise cause and restore the foundation to human knowledge that concern about an epistemic crisis uncritically implies. As with any other exploitation of the surplus value of network capital, we can advertise porn and pills.
Or we can use these essays to tackle the actual epistemic crisis within academia, which isn’t the need to teach students and faculty critical thinking and proper research and citation practices. Rather it is how to present knowledge that is suppressed by the academy and its systems of epistemic authority. Knowledge underwritten by racialised, gendered, colonised, classed, and disabled epistemologies rather than by citational gestures that fail as soon as the surplus value of the aesthetics of their authority is exploited by LLMs. If the academy will not stop working to contain these through its successful operation, then they can exploit its failure in order to open a breach that they can enter into where they are otherwise excluded.
Slopsquatting the episteme for epistemic justice.