Sign in Subscribe
4 min read Almost Entirely Human

Episode 34 - Bob Ross is back and specializes in… Excel?

Sora2 means that deepfakes are easier than ever, making the possibility of misinformation more widespread too. Sure, making Bob Ross paint Excel is fun, but if someone can make Bob Ross talk about Excel, imagine what they could make YOU talk about, even if you don’t want to.
Episode 34 - Bob Ross is back and specializes in… Excel?
A portion of a GPT generated image that interprets this post. It contains a dog in a parachute harness that maybe has a wing just doesn't look exactly right, but for a cartoon sketch is exactly what I wanted for this post.

Prologue

A few weeks ago, in Episode 30, I talked about not being a secret cyborg. Then last week, I mentioned in the news that OpenAI released a new video generator called Sora2. I haven’t yet used Sora2 myself, but I have seen the results, and they are pretty scary (also impressive).

📽️
This post has several videos in it. If you open this post on the web, you can watch them all inline without visiting external sites.

[An AI generated video of Bob Ross talking about the best way to use colors in Excel, it is pretty amusing].

TL;DR - Sora2 means that deepfakes are easier than ever, making the possibility of misinformation more widespread too. Sure, making Bob Ross paint Excel is fun, but if someone can make Bob Ross talk about Excel, imagine what they could make YOU talk about, even if you don’t want to.

Fakealogue

Here is a fun example where a chef critiques alien food:

Chef critiques alien food on MasterChef, Sora 2 (AI Generated, not real, not even a little bit)

Sure, I don’t think anyone is going to believe that (well, not too many people, anyway).

Here is another one of a golden retriever riding tandem while skydiving:

A dog parachuting, GoPro footage, Sora 2, AI Generated. Not real, not even one bark.

[Editor's note: please do not try this at home.]

Not Perfection, But Close

If you watched any of these, you’ll have noticed little issues here and there, like the dog barking, but his jaw doesn’t appear to move. These AI-generated videos are not perfect, but they are so close that while scrolling on social media, you might not notice that anything is off unless you pause to watch it a few times and look for inconsistencies. That’s the problem, just like with the rabbits on the trampoline, you have to pay close attention to know it is fake.

The Watermark

All of the videos have a jumping “Sora” watermark on them, both hard to remove and reasonably obvious—if you know what it means.

I suspect that most people don’t know what “Sora” is, and don’t know that it is an AI-powered video generator capable of making just about anything you can describe with words.

A screen capture of the Bob Ross video from above circling the Sora logo that you can see in all Sora generated videos.

OpenAI Takes Action

This tool has only been out for about a week, and it is already making news headlines:

So, OpenAI stepped in to try to make it harder to generate copyrighted material, turning up the sensitivity on the moderation process. At one point, the sensitivity was so high it was hard to make anything!

🙆
They are limiting the generation of videos that infringe on copyrights. I wonder how they do that? Probably another AI model.

I guess I need to go copyright my face.

Politics

We don’t all agree on everything, and I think that’s okay. But I think one thing we can agree on is that the truth is important. So when scrolling on social media, look out for that moving Sora logo… be sure what you are believing is true.

What do we do?

  1. Don’t be a secret cyborg — If you post AI generated content, disclose that!
  2. If you see misleading content on the internet that is AI generated, add a comment!
  3. Let people know about the “Sora” watermark. Don’t assume everyone knows!
  4. Share this post with one person! (Okay, yes, that is also self serving, but seriously, at least talk to your friends about deepfakes)

Newsologue

Epilogue

This post didn't get an AI review, I wrote it. Holly editied it, then I published it.

I guess I was feeling a little anit-ai when I wrote this, but rest assured, I am still using it!

In fact, we played a fun game at dymaptic this week where we all learned a lot about how AI models work and how they can fail in production. Stay tuned for more!

Published by:

Christopher Moravec

You might also like...

03
Oct
Episode 33 - Trust But Verif-AI

Episode 33 - Trust But Verif-AI

Two AIs are better than one - If we use multiple models, we can obtain different interpretations that we combine to yield a more reliable answer.
5 min read
26
Sep
Episode 32 - Building a Fantasy Map Machine

Episode 32 - Building a Fantasy Map Machine

I built a system to help us collect information at conference booths in a fun and engaging way—building you a custom fantasy map.
3 min read
19
Sep
Episode 31 - Why do models make stuff up?

Episode 31 - Why do models make stuff up?

The term of art for AI is called hallucinations, but I prefer to stick with Make Shit Up. The training incentives are flawed; we have been training models to hallucinate because they score higher on tests by guessing incorrectly than by admitting they don’t know.
5 min read
12
Sep
Episode 30 - Don’t be a Secret Cyborg

Episode 30 - Don’t be a Secret Cyborg

We need a culture where it is okay to admit to using an AI and celebrate it, all while keeping the human component in mind. Even if AI models don’t improve beyond their current state, we already face numerous issues, anxieties, and superpowers!
7 min read
05
Sep
Episode 29 - AIs don’t learn, but they sure look like they do

Episode 29 - AIs don’t learn, but they sure look like they do

LLMs do not learn on the fly, their training is compute-intensive and cannot currently be done dynamically. Instead, they can recall messages from previous conversations, making them appear to have learned things about you.
6 min read

Member discussion