Jeremy Carrasco makes videos on TikTok under the handle @showtoolsai advocating for AI video literacy and pointing out that you can tell if a video on your feed was generated by AI.
Alsa Chan, host:
AI is everywhere, right? You're probably seeing this phenomenon all over your social media feeds. Remember the cute bunny jumping on the trampoline? Fake. And now there are so many of these fake videos that there is a kind of AI video critic.
(SOUNDBITE OF ARCHIVED RECORDING)
JEREMY CARRASCO: This video uses a different AI model. I think this is a Chinese model of Kling because there are Chinese letters on the ceiling. And when he runs his hand through your hair, his little finger sticks perfectly there. They sell research chemicals and skin care products. This can't be reality.
CHANG: Jeremy Carrasco. He is a video producer and has spent years screening real videos for errors. Now he's keeping a close eye on AI. And TikTok and Instagram are teaching people how to spot a fake. Carrasco says improved AI models are making it much harder to spot fakes. Even if you lose a limb or finger, you are no longer dead. However, the model is still not perfect.
Carrasco: For example, there are AI scams that ask people for money while they are sitting in a hospital. But if you send these images to your doctor, you'll likely find that the device is either completely wrong or half-finished. If you watch videos of rock climbers, you'll see that ropes are either dangerous or tied to nothing.
Chan: (laughs).
Carrasco: And as a casual observer, you might not know that. But the good news is that you have other superpowers and expertise in other areas that may be helpful in other situations as well.
CHANG: So even if the scene seems real, it could be fake by your own logic. He says that even things like camera angles can help you think critically about what you're seeing.
Carrasco: Of course. The main difference between these videos and real videos is that there is no real person behind the camera. As a result, camera movements often don't make sense. And we just need to take a step back and ask who filmed it and why. Numerous videos have been posted online of a dog rescuing a baby from a falling shelf with a perfectly placed camera.
Chan: Yes.
Carrasco: You know, that wouldn't happen in real life. And if someone has a cell phone and it looks like the baby is sticking something into the outlet, that person will put a film on it to help.
CHANG: (Laughter) Yes, that's right. Or something like that shot of a cat jumping onto someone's bed and grabbing a snake in the middle of the night. For example, why would you need to pan yourself on a surveillance camera while you're sleeping at night?
Carrasco: That's right. I mean, they're not all that obvious. But if you can find these obvious ones, I'll tell you – pet videos are great because they appear in all of our feeds. If you can capture those obvious problems and learn from them from other details, you're on your way to becoming better at spotting them overall.
CHANG: Well, I saw that TikTok has a policy that requires creators to label all AI-generated content, including realistic images, audio, and video. That's the policy. How reliable are those policies, and how widespread are such policies among social media companies these days?
Carrasco: Very unreliable. And they have some detection power. However, the detection seems to be based on descriptions and hashtags rather than actually analyzing the video. As far as I know, no social media platform takes detection or labeling seriously.
CHANG: Okay. Okay, let's move on to the next question. Some of your videos talk about technical watermarks that AI generators embed in content. But are they reliable indicators? That is, what do these watermarks look like to ordinary consumers encountering these videos?
Carrasco: Well, there are visible watermarks and invisible watermarks. So AI companies like Google are trying to enter invisible watermarks embedded into pixels. But if you give Google an AI-generated photo that wasn't created by Google, Google can't tell whether it's AI or not.
Chan: Yes.
CARRASCO: So, right now, there's no good, reliable, centralized solution. But there are definitely some promising things that will hopefully be implemented soon.
CHANG: So, God, as I'm talking to you right now and thinking about all the methods we use to spot fabrications, I know that what I feel like I'm getting lost in here is just the spontaneous joy that I would feel when I come across an interesting video.
Carrasco: Yeah.
CHANG: So I'm wondering, how do you think that hurts all the video creators who are making authentic work?
Carrasco: I agree with you. This is the saddest thing. Because the best way to discover AI videos is to embrace that strange feeling you may have felt…
Chan: Yes.
Carrasco: …So, listen, wait, is that because it's AI now?
Chan: Yes.
Carrasco: And what I'm saying is that there are still a lot of real creators out there who can give you that kind of joy. And just by following them, you know you can trust them not to cheat you. Because a lot of real creators aren't actually interested in making AI videos. And AI accounts actually exist to create AI. Therefore, I don't think there is a need to eliminate that mysterious feeling. All you need to do is know where to look for it, and don't rely on the social media platform's algorithm to simply show you it.
CHANG: I'm Jeremy Carrasco, an AI literacy educator. Thank you for your valuable feedback.
Carrasco: Thank you, Ailsa.
Copyright © 2025 NPR. Unauthorized reproduction is prohibited. For more information, please visit our website's Terms of Use and Permissions page at www.npr.org.
Accuracy and availability of NPR transcripts may vary. Transcript text may be revised to correct errors or match audio updates. Audio on npr.org may be edited from its original broadcast or after publication. The reliable recording of NPR's programming is the audio recording.
