I got a golden ticket to the slop factory (thoughts on the Sora app) (kkukshtel.substack.com)

I got early access to Sora, a newly launched AI video app that treats short videos like remixable, latent-space artifacts rather than fixed creator outputs. Sora’s feed surfaces horizontal stacks of variations on the same prompt so you swipe through many permutations of a source video; each remix is mathematically close to its reference in the model’s latent space, making works simultaneously derivative and “original.” The model impresses technically—coherent edits, strong audio, and context-aware dialogue—while the product deliberately centers prompts and communal remixing over individual creators, effectively normalizing “author-death.” That design has big implications: it reframes creative discovery around prompts and trend-chains, accelerates viral imitation, and raises pragmatic and ethical challenges (widespread deepfakes, celebrity cameos, and lowbrow misuse like parodying MLK). UX and tooling are weak—no prompt refinement tools, no multiple generations, no seed reuse—and generation speed is slow compared with image tools, limiting iterative exploration. For AI/ML practitioners this is a GPT-3.5-style inflection point: Sora shows how model quality + community can be a moat, but also that success will hinge on moderation, faster/cheaper inference, and richer “model UX” beyond a prompt box.