Latest / Logosofia no Dia a Dia
Beyond Common Crawl: The Datasets Shaping AI
Everyone talks about Common Crawl and The Pile, but they're just the bulk ingredients in the AI pantry. This episode explores the selective, purpose-built datasets that truly shape model behavior. We dig into the challenges of building multilingual models, how labs like Meta and Alibaba update world knowledge in point releases without breaking what models already know, and why almost no one trains specialized medical models from scratch. Episode 5116 — myweirdprompts.com/n/5116 (permanent short link: myweirdprompts.com/908305)
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.