How to dirty your data (workshop)

25 places available - SAVE YOUR SPOT

So-called AI models are largely fed with content scraped from your, my and whatever available websites. These models cannot self-clean like cats, so how come they are not immediately poisoned by intentionally placed prompt injections? During the workshop, we will first examine how extracted data is prepared to be used for AI training, following “good enough” rules (a.k.a heuristics) devised by software engineers. Then we will develop a set of anti-heuristic heuristics to make our data too dirty to be trained on.

jiawen uffline exists as a user most of the time in their life, is watched by satellites, cell towers, their computer, phone, browsers, door cams, tooth brush and their government. Having little agency both technologically and politically, jiawen seeks for possibilities in internet cracks for queering the given identity of a user. Their work counter-narrates and intervenes the only-path-forward narrative of technological stability and purity established by big techs.