Brief notes
- These days are very long! Starting at 5am PST, continuing through the evening
- Control theory (like for controlling airplanes), but for exploring gradients! Handling second-order issues like momentum
- Anna Hoffman: Something beautiful had been ruined forever
- AI ethics often starts from an abstract idealized world view, instead of being grounded in the world as it is today
- (the ruined beauty was the innocence about the goodness of the world)
- conceiving of AI as forever growing and always bettering can ignore that problems along the way are related and a direct product of AI itself
- attempts to make algorithms more fair can use a reflection of existing social structures (two genders, for instance, which is fewer than the California DMV)
- tough to prevent future gaydar facial recognition studies, in the real world: more effective to focus on current oppression
- bioethics has interesting parallels, but it is also very much a product of historical discourse, and AI is much newer
- What are we going to do about Computer Vision?, a panel in Queer in AI
- Critical work is necessary! And critical work does not need to end on a note of resolution 😅
- Participation isn’t enough: everyone on all sides can intertwine with oppressive systems
- Sometimes the best participation is refusal!
- The Luddites were skilled machinists who were told that they were getting fired and replaced by machines, so they smashed the machines
- Low-data low-compute is always alluring, obvi
- StyleGAN with data augmentation has good results on only a thousand images! Especially after finetuning from FFHQ! Collecting 70k faces is challening, but 1k faces is few enough to come from an art museum
- To read: The Eureka Effect: The Art and Logic of Breakthrough Thinking
- To explore: TVM
- Frightening: identifying personal genomes by surname inference