Sora: Infinite Meaninglessness (thelastwave.substack.com)

OpenAI has quietly launched Sora, an invite-only iOS app that serves a TikTok-style feed of short videos entirely synthesized by its Sora 2 model. A standout “cameo” feature lets users upload faces to generate startlingly consistent deepfakes — friends, public figures (including Sam Altman), and copyrighted characters — placed in uncanny, attention-grabbing scenarios (think Pikachu staging a robbery). OpenAI pitches this as a “ChatGPT for creativity,” emphasizing fun and fast idea-to-result workflows and character consistency, while early tests already show the format’s memetic potency. For the AI/ML community this is a flashpoint: Sora compresses video synthesis, user prompting, and recommender-driven optimization into a feedback loop that can flood feeds with “never-events” optimized for engagement. That raises technical and societal concerns — provenance and watermarking of synthetic media, robust deepfake detection, recommender-auditability, dataset contamination from platform-spread synthetic content, and studying long-term cognitive effects of prolonged exposure to synthetic semiosis. Practically, researchers and practitioners must prioritize attribution standards, adversarial detection, and alignment of recommendation systems to avoid amplifying addictive or misleading artifacts; policymakers and platforms will also face pressure to govern how generated media is labeled and distributed.