Filtered by category: e-Literate, tag: language models × Clear
Literally Nobody Understands AI. That’s bad.
Matched: AI
AI remains poorly understood even by experts — no widely accepted scientific theory explains how it works. Models trained only on text somehow learn grammar, context, and even internal representations like chessboard states, yet fail at simple tasks like tracking who said what. Without falsifiable explanations, AI failures are unpredictable and hard to fix, posing real risks in applications like tutoring. Research into smaller, more efficient models is beginning to close this gap.
AI Mindscape Prompting
Matched: AI
A conversation between educator Michael and Gemini AI explores how large language models actually work. Using the "apples and oranges" idiom as a framework, they discuss how AI processing resembles human cognition functionally—activating relevant knowledge based on context—while differing fundamentally in mechanism. Key insights include how prompting works as collaborative "mindscape" building rather than magic incantations, and how context windows enact distributed cognition and social constructionism concretely.