Human Oversight Makes AI Media Automation Work (Full Transcript)
[00:00:00] Speaker 1: You can actually get Underlord to do pretty meaningfully complex visual edit and media assembly tasks. You just have to give it very explicit instructions, and it needs to be replicable, right? Like the benefit of going through this whole quest of creating a skill and turning it into an automation is if you tend to do it a lot. When you give explicit instructions, the eyes won't be doing those sorts of tasks. And even if you have a really great prompt, I think we still expect for humans to be in the loop. There might be scenarios one day in the world where you just got a content machine, you don't touch, it goes, it posts, and the world is drenched in slop and we all die a miserable death. But until that point, we expect humans to continuously refine. And I think the scenario we'll show today illustrates why humans in the loop make sense. It's just that instead of a human starting the entire process, the first time a human will touch a social clip in our workflow, it will be once the clip is already made.
- AI can perform meaningful visual editing and media assembly with explicit, replicable instructions.
- Automation delivers the most value for recurring workflows.
- Human review remains important for refining AI-generated content.
- The workflow shifts human effort from starting production to reviewing an already-created social clip.