Great panel at #DH2025 and it's great to hear more about AI resistance & refusal, giving consideration to whether our methods serve the communities we work with.
Though this ex. does raise questions for me about the impacts of the absence in training data if scholarship is closed.
@gworthey.bsky.socialFirst @filipa-calado.bsky.social addresses language, copyright, Fair Use defenses, bias, etc. in AI.
Filipa trained two GPT-2 models, one on Heritage Foundation-promoted texts, and the other on ACLU-promoted texts. Hilarity ensues (along w/less-funny stuff).
#DH2025
Slides: bit.ly/dh2025-open-...
bit.ly
dh2025 open data
‘Original Expression’ to ‘Aggregate Forms’: Language and Copyright in the Shadow of AI Filipa Calado Assistant Professor Pratt Institute, School of Information Link to slides: bit.ly/dh2025-open-data