Open source projects have long been THE critical dependencies in our digital infrastructure, as xkcd illustrated. This talk will softly trace three of the soft dependencies in heuristic filtering data for pre-training large models. What formed the five categories for classifying NSFW images that later was expended to NSFTM ( Not-Safe-For-Training-Model), where is the blocklist every dataset is using for blocking bad content?
Zooming out, do dependencies have to be so arbitrary? Can we imagine a softer, decentralised interdependency? In this talk, we will also discuss the possibilities of trans*feminism's imaginary of dependency.
