John Feminella 🌠

@jxf
564 Followers
75 Following
502 Posts
⏪: Two Sigma, Thoughtworks, Pivotal, VMware, EY
🥰: bits, bots, bucks, blocks
👤: he/him
🦋: Bluesky @jxf.me
In this house, we obtain our facts only from America's finest news source.
Tired: Elf on a Shelf
Wired: Elf with a Ring
I'm already seeing some users refer to this flavor of content as "sloptok", a fun portmanteau of "AI slop" and "TikTok". Probably doesn't bode well for the quality of informational content we're about to experience over the next few years.

I think that to the casual YouTube Shorts, TikTok, or Douyin viewer, or to the general untrained eye, it would likely be challenging to identify this video as clearly AI after one viewing.

To me it is clear after a couple of viewings that it's AI-generated (note the street signs, traffic light distortion, face masking, etc.). But even though I've seen a lot of AI content, I'm not sure I could nail it 100%.

While objective evaluation of videos is much harder than objective evaluation of textual answers, it's nevertheless very clearly a capability leap over the earlier generation.

Prompt:

---
Scene: A NYC street interviewer asks several passersby if they know that they're just the result of a generative AI model's output.
Direction: The passersby act shocked, in denial, or in disbelief when they learn this.
Camera: We cut between each interviewee over the course of the clip.

OpenAI recently released Sora 2, their new video model. I had access to preview it this past week.

Just to give you a flavor of the general capabilities, here's a zero-shot 15-second video I made (no editing, no revisions).

ln(😅) = 💧ln(😄)
This deal makes a lot of sense given that a large part of the American cereal aisle basically *is* candy.
What if every time you went to thispersondoesnotexist.com, it wasn't actually using generative AI but rather causing a real person's instantaneous and simultaneous erasure from this timeline?