Can OpenAI build a social network? (maxread.substack.com)

OpenAI’s new app Sora brings state-of-the-art text-to-video generation into a TikTok-like social feed: users type prompts, wait ~5–10 minutes for a generated clip, and can post to “For You,” “Latest,” or follower feeds. A standout feature is “cameos,” where you upload a short clip of yourself to create a promptable likeness others can use (with permission); high-profile figures like Sam Altman have allowed open use, accelerating viral deepfakes. OpenAI enforces some guardrails (blocking certain sexual scenarios and limiting some living-person generation), but Sora still enables easy videos of deceased public figures and politically charged fabrications, and its relatively loose safety measures have amplified both creative and risky uses. Sora matters because it signals a strategic shift: OpenAI is packaging generative models as a social platform rather than a chatbot tool, potentially leaning toward engagement-driven business models (ads, network effects) akin to Meta’s. That raises three technical and industry implications: (1) verification and trust—cheap, believable video deepfakes further erode evidentiary value of footage; (2) moderation and safety—current guardrails are patchy and scaling is hard; (3) economics—unlike hosting static user content, every Sora clip incurs nontrivial inference cost, challenging the favorable unit economics of social platforms. Sora’s viral success shows demand, but its technical costs and safety trade-offs make it unclear whether this is the start of a new social paradigm or a costly experiment.