In Defense of Generative Video Accelerationism (www.thediff.co)

The piece argues provocatively that a future in which generative video is so ubiquitous that viewers must assume any clip may be fake is not only inevitable but potentially salutary. Drawing historical parallels to printing and the social shift around telephony, the author says technologies that initially corrupt trust eventually force new norms and verification practices. Video is uniquely pernicious because it mimics lived experience and breeds overconfidence and selection bias: cheap, viral clips (real or synthetic) can make fringe events feel ubiquitous. Generative models will therefore amplify sensational, belief-confirming content, outcompeting ordinary footage and accelerating misinformation unless social and technical responses evolve. For the AI/ML community this is a call to action: expect pressure for provenance, chain-of-custody, robust watermarking, dataset transparency, and automated detection/attribution tools as cloud-backed cameras and content platforms create verifiable metadata. The essay’s economics (cheaper generation → more low-quality output; “Alchian–Allen” style selection effects) and sociotechnical dynamics imply model developers must balance innovation with stewardship—building provenance standards, demonstrable watermarks, and forensic detectors while preparing for adversarial uses. Ultimately, the author predicts an equilibrium where “I saw a video” is no longer persuasive without provenance, improving collective epistemic hygiene even as the short-term landscape grows messier.