I couldn't wait to try out Sora, the new AI video generator. Like many, I uploaded my images and voice, eager to see what kind of wild videos I could create. My first thought was to create a funny video of my AI self getting pelted with paintballs by a bunch of elderly folks. It sounds silly, right? However, after playing around with it, I realized there's more to Sora than meets the eye.

Sora is powered by a pretty advanced AI model, Sora 2, and the videos it produces can range from mundane to bizarre. It feels almost wrong to use, even if you can't quite put your finger on why. If you've been playing with Sora too, here's the lowdown on some of the less-obvious downsides.

The Energy Cost

Making just one Sora video eats up a surprising amount of electricity. Estimates suggest it could be around 90 watt-hours! To put that in perspective, some sources say it's like running a 65-inch TV for nearly 40 minutes. OpenAI hasn't released the exact figures, so these numbers are based on educated guesses and similar AI models. The actual training of Sora consumed a HUGE amount of electricity. So, when you use Sora, you're indirectly contributing to that massive energy footprint.

Think of it like this: using any cloud-based AI model is similar to ordering surf and turf at a restaurant. While the cow is already dead, the lobster is still alive until you place your order.

Water Usage

Data centers, where AI models like Sora run, require lots of water for cooling. A single video generation might use more water than you'd expect. Estimates suggest it could be around 0.17 gallons, which is more than a bottle of soda! So, generating videos isn't exactly a water-wise decision.

Privacy Concerns

Sora's Cameo feature, which lets you create videos using your likeness, has privacy settings, but you need to be aware of them. You can control who can use your Cameo and even set rules about how you should appear in videos, including what you should never be shown doing. This is a great way to protect yourself, however, be careful!

Content Guardrails and the Unexpected

While Sora has content filters, they're not perfect. In fact, OpenAI admits that offensive content can sometimes slip through. In my own testing, I found that the filters generally worked, but some objectionable content was surprisingly unfiltered. I even managed to generate videos of a person interacting with poop. Yes, you read that right. Surprisingly, previous AI models had measures in place to prevent that kind of thing, but it seems to be missing in Sora. I don't see it as a scandal, but its pretty nasty.

The Spread of Misinformation

Sora can also be used to create incredibly realistic hoaxes. For example, a viral video showed an AI-generated Donald Trump supposedly talking about the Epstein files. While the creator claimed it was for artistic purposes, some people actually believed it was real. If you post videos publicly on Sora, others can download them and do whatever they want with them, including spreading misinformation. Once you put something out there, you lose control of it.