The coming tide of deepfakes

Well, this stuff has been brewing for years and it’s terrifying. We’re on the cusp of entering a world where you no longer will have any idea whether what you see online was created by a human or an AI. For that matter, it’s hard to know how media will spin, what “news” organizations will create to drive ratings, etc.

Similarly, this entire subreddit (posts and comments) is created by bots running the OpenAI GPT-2 model:

https://www.reddit.com/r/SubSimulatorGPT2/

Here’s a subreddit that tries to collect some of the best safe-for-work deepfakes:

https://www.reddit.com/r/SFWdeepfakes/

Some of this stuff is funny:

Some of it is racist:

Etc…

Here you can generate your own text.

https://talktotransformer.com/

For example, I fed it text from this qt3 article:

Microsoft is relaunching the IntelliMouse as a gaming peripheral. When the first IntelliMouse appeared in 1996, PC users lauded its clean lines, ergonomics, and introduction of the scroll wheel. It quickly became the standard by which other mice were compared. In 1999, it became one of the first commercially available optical mice thanks to “IntelliEye” technology. Throughout the years, Microsoft has sold the venerable mouse to hardcore fans, but now they are offering a jazzy gaming version.

It completes the text…

‏In the first year,‏ a total of 80 to 100 mice (approximately 40% of the industry) were purchased during the first quarter of this year from our dealers through our dealer online store“.“ We’ve also offered an Intellicon mouse designed for older games for the last 6 years.

“To date, we’ve sold a total of 400 different mice.“We want to emphasize that if you’re thinking “this is too pricey for me!”“,“ then this is a good opportunity to try IntelliMouse.“

We are very honored to announce that as part of our ongoing initiative to promote Intellimove, we will offer the entire IntelliMouse line, including the first IntelliMouse for older games for a discounted price.“

Here’s another example… I’ll feed it this recipe text:

  1. Preheat oven to 325 degrees F (165 degrees C). Grease and flour a Bundt pan.
  2. Cream 1/2 cup plus 2 tablespoons butter and 1 1/2 cups sugar until fluffy. Add eggs one at a time beating after each addition. Blend in the lemon peel.
  3. In a separate bowl, mix flour, salt, soda and baking powder. Add flour mixture alternately with buttermilk to creamed butter mixture. Add lemon extract and raisins.
  4. Bake at 325 degrees F (165 degrees C) for 50 minutes cool 5 minutes, then turn out onto serving plate. Prick hot cake with skewer or fork and pour on lemon topping.
  5. Combine 1/3 cup sugar, 1/3 cup butter and water in a saucepan and heat until butter melts. Add lemon juice . Spoon over hot cake.

And the continuation is:

Meanwhile, preheat oven to 350 degrees F (180 degrees C). Line a 9×9 pan with parchment paper. Place butter on parchment-lined baking pan, lined, then roll out all sides.

Bake 10 minutes, then flip to one side for second baking and allow to cool for 10 minutes to completely cool.

While the cakes are cooling, prepare filling. In a medium bowl you will combine 2/3 cup heavy whipping cream with 1 cup sugar.


Deepfakes are interesting and really just emerging into the territory of believable. It is going to be interesting to see how they are used and abused in coming month/years.

Honestly I don’t think this really changes things. The internet has long been full of sites that will use selective or out of context quotes/pictures/videos to heavily spin stories or just flat fabricate them. So in that sense, the advent of fake audio/video is just a continuation of the status quo that most things seen on the internet need to be taken with a large grain of salt.

How is the Mike Tyson video racist?

Maybe it’s not? I don’t know. Seems like they’re taking advantage of stereotypical black vernacular and Mike Tyson to try to make something funny?

I mean, honestly, I don’t know. What does it mean to be racist with something completely fake? Or partially fake?

The nsfw stuff is pretty mad. There’s a gif of Jordan Petersen fucking a hotdog roll doing the rounds on the chans. I presume its a fake lol

The video they used to fake is an old meme that was popular in the mid to late 2000s. I don’t think it was a racist thing to put Mike Tyson in it.

But yeah, it’s a very interesting topic and I appreciate you starting the discussion.

I remember having a conversation with a colleague about 20 years ago. The discussion was around the definition of “art” and artistic integrity. One of the points was: when there comes a day that you can create completely fake but entirely realistic child pornography, is that still considered art? Or should it be illegal? How would the SCOTUS rule on the 1st amendment rights of an artist creating fake child porn?

There are a lot of interesting legal angles to this stuff. For instance, Zuckerberg has said he won’t ban deep fakes on Instagram. What does Kim Kardashian do when somebody starts creating deepfakes of her talking nonsense about conspiracy theories? Can she sue for libel/slander? Does she own the right to images that look like her, even though they aren’t? How will it possibly damage her reputation and harm her career or family?

That fake subreddit is awesome and fascinating.

Well my old adage of ‘if Zuckerberg thinks it’s a good idea, it’s probably a terrible one’ still rings true.

I had this old idea I used to pop out at parties as a conversation starter. “What if some big budget special effects company decided to completely screw you over by using all of its resources to create a fake video of you committing some capital crime?” The discussion would always wind up centered around the fact that the amount of money needed to do this would be prohibitive and you’d have to somehow get everyone at the effects company to keep the secret. Now? Those factors are gone.

This will only cause citizens and consumers to be more watchful and less trusting in what they see, strengthening their judgment out of necessity.

(muffled laughter)

Legally, I can’t see any difference between falsely reporting that someone did or said X bad thing, and creating a video which falsely shows that person saying or doing X bad thing. In either case I think it would be actionable. Assuming you can find the originator, of course.

The tricky thing is the legal treatment of those who share it. Some will share it knowing it’s a fake (they do that now), while some will no doubt be ignorant. Since you can’t easily prove knowledge, I doubt there is any legal recourse against them.

When the “Bigger Picture” is what we’re watching, the news will appear much more like an Internet chat room when we’re online, and we will see what’s on the internet being reported on by different people all over the world. We will hear rumors, we will see what looks good in the press, we will hear about other topics, we will have no idea who or what news is real.
The truth is always bigger than the mainstream.
The reality is when I go online and read a news article, it is something in my newsfeed but I rarely see it. This may not be a good thing. When I go online and look at a video, even that may not be what is shown up top in my newsfeed.

Slightly edited output, but talktotransformer is on to something. here. :P
Knowing a bit about it, language will probably be the hardest to fake because there a ton on contextual implied meaning in words and phrases, but it’s also the easiest to touch up by a human. Either way, this is really going to be an interesting time to go through, assuming we get through it.

Your identity is your property and should be protected by, like, laws and stuff. Deep fakes are identity theft and should be prosecuted as such.

If Govs. Schwarzenegger and Ventura give their permission to deep fake a fight scene in Running Man 2 that’s fine. If some dude does it on instagram and says he caught Arnold fighting Jesse at a bar, he should be subject to a lawsuit and criminal prosecution.

This is my opinion, but it seems pretty clear. Make it a crime to do this to someone without their permission.

Check out the video on this page of Trump and Putin singing Lennon’s Imagine.

My experimental deepfake forum entity AI “Clay” is becoming quite precocious.

I recently saw a news story where they talked to a guy from Stanford or Berkley who explained how this worked and gave a demonstration. I can see fake videos exploding in the run up to the election. That a large percentage of the viewers won’t care about the authenticity of the videos is the scary part.

Edit: Nevermind! I just reread the OP and saw that it really was a fake subreddit!

The scarier thing to me is that it’s an assault on the very concept of truth. It’s not that some people will believe the fake videos, it’s that the waters get muddied to the point where people are too cynical to believe anything. This is a trick that authoritarians use and is kind of the basis of Trump’s fake news mantra. Most rational people can dismiss the idea when it comes from Trumps mouth but when deep fakes really start making their presence felt, then many Americans will start being skeptical of pretty much everything they see and hear.