I'm not going to be a vigilante, but I think it's time to assume that everything you see online is fake.
Of course, the internet is full of content created by real people (included in this article). However, the media generated by AI has become so realistic that it is almost a disadvantage to speculating that content that you have scrolled in the past in your feed is legal.
Don't skip this article because you know what AI content looks like. If you know what you're looking for, it's easy to find the current ones that algorithms provide to your social media feed. However, even if you can identify the AI slop in seconds that hit your eyeballs, you need to know that you are not ready for the next wave of video generated by AI. That wave isn't just about the path. It's already here.
AI content is already cheating on people
Most of us are keenly aware of the appearance of “AI video.” This “tragic” video of a cat's parents abandoning them from a burning plane to save a kitten is obvious to most people who watch it. Perhaps you know that Trump doesn't work at this construction site, and you most certainly can understand that the cat farmers of this family are actually AI-generated.
But there is a video that is less obvious. This is not particularly true for us who don't quite match up with AI and technology. You may know that this video of a baby dancing in a circle is AI, but many people in the comments did not (assuming they are not bots either). Also, this pet family could tell you that they didn't actually see the bird inspecting the toy crocodile, but again, they couldn't. And there's no end American talent Featuring “realistic” yet impossible visuals, the video captures the hearts of hundreds of thousands, if not millions. (I cry.)
However, I am not writing this work today. Because I'm worried about how many tricks these “believe” AI videos are method Too many people think they are real. I'm worried about it, but those worries are pale compared to my new horror.
So far, most of the AI videos that have taken over social media feeds have relied primarily on visuals and background sounds to sell suspicious reliability. You'll notice that none of the characters in these videos actually talk about. If so, lip movements are not synchronized, usually without the robot's voice and are quickly off-patched. It was easy for AI creators to focus on people and animals realism in their videos. I hope that the baby dancing with the lion is surprised enough not to think, “This is bullshit.”
Even the Openai Sora video model, which I shocked at its quality last February, worked from its realistic visuals. The video of a woman “shooting” her reflection from the train window is too realistic for comfort, but Sora didn't let out a fully rendered conversation. If you see such a scene in your feed, of course, I think it's a real video, or at least something that's generated by humans.
AI video is about to change completely
Something happened this week, just making me more pessimistic about the future of truth on the internet. At this week's Google I/O event, Google announced its latest AI video model, the VEO 3. Like other competitive models, VEO 3 can generate very realistic sequences. Certainly not great, nothing Really There's new there.
But VEO 3 can not only generate videos that may trick you into thinking about the reality. Veo3 can also generate audio along the video. This includes not only sound effects, but also dialogue.
To demonstrate the audio/video capabilities of the Veo 3, Google showed off a clip of an old sea sailor in the sea. The quality of the video is sharp and realistic, and the words that the man speaks are synchronized with the movement of his lips. Of course, I realize that video is AI, so I know that it's a habit of providing games (in my eyes, this looks like a higher quality animation than live action shots), but I'm sure this video will be fooled. Many Fake AGT video fans.
But even this clip didn't affect my new fear. It was a video that users started making when they got Veo 3. Petapixel has a great roundup of some of the “best” Veo 3 videos people have ever made, but highlights some of the things that should scare you the most.
This clip shows that the streamer is playing Fortnite. Everything, including game footage, was generated with Google's AI:
This tweet is currently unavailable. It may have been loaded or deleted.
The clip presents three concerts that never happened, featuring non-existent musicians and crowds. The music isn't good, but that's not the point. From vocals to instrumentals, music was completely generated by AI and synchronized with lips, drums, guitar and strings.
What do you think so far?
This tweet is currently unavailable. It may have been loaded or deleted.
But this clip definitely should sound an alarm for each of us. Someone generated fake videos of fake car shows and featured fake interviews with fake participants. It's far from perfect, but the quirks of AI are completely hidden by the surface-level realism here. Not only will AI Got Talent fans buy this, I I'll buy this, especially if I wasn't looking for it:
This tweet is currently unavailable. It may have been loaded or deleted.
It's visual. It's a dialogue. It's a crowd. It's lighting. That's a candid laugh at “mistakes.” It's the sound of the microphone being “clashed.” Congratulations on your realising the dialogue. It often makes no sense, or that people in the background ignore the laws of physics. You won't notice it when you get a hit mid-scroll on Tiktok or Instagram.
Even the VEO 2, which is less powerful than the VEO 3, offers tools for realism, including the ability to tell you how you want your camera to work. Additionally, both models are available in Google's AI video editor Flow. Creators now have the ability to generate very realistic AI content. This just feels like it was filmed face to face and the technology is getting better.
This tweet is currently unavailable. It may have been loaded or deleted.
Google's best AI video generator tool costs $250 a month through the new AI Ultra subscription plan. It's expensive, but it's not out of reach for anyone interested in creating AI-generated content. However, the AI Pro, a $20/month plan, still comes with VEO 2 and Flow Access. The rate limits are lower, but I'm not shocked to see realistic slops coming out of these limits either.
It's time to become a full-time skeptic
None of these techniques are perfect. Everything that Veo 3 spits out is not here to tell you that it cannot be distinguished from the actual content or that the video is not in regular AI. In fact, there is clearly something about the training data for Veo 3. As a report from 404 Media, models continuously generate the same strange “dad jokes” whenever they ask to play a generation of comedians.
What I'm saying is when I turn on the bullshit detector and keep it active full time. When involved in videos on the internet (especially in short form of algorithmic clips), the content works more securely assuming it is fake from the jump, requiring evidence beyond the reasonable doubt that what you are looking at was not generated with a simple prompt and a $250 budget. That feels extreme, but after watching this week, I have not really seen another way to engage with this content in future involvement.
We are in scary territory now. Today it's a demo of musicians and streamers. Tomorrow, it's a politician saying things they didn't. A suspect who committed the crime he was accused of. A “reporter” where you lie through the “news.”
I hope this is as good as technology can get. We hope that AI companies run out of training data to improve their models and that governments take some action to regulate the technology. But I saw US Republicans pass bills that include a ban on AI regulations that are enforced in 10 states. yearI am quite pessimistic about the latter point.
Perhaps this technology will be better used to use zero guard rails to ensure that it goes safely. I'm wondering how many politicians who voted on that bill this week after watching an AI-generated video on their phones and had no idea about it.
