Chan­ging the World with Com­pu­ter Visi­on: How Noema use ge­nera­tive AI mo­dels to ma­ke re­al­-world i­mpacts?

How does Noema use generative AI models to make real-world impacts?

Noema's VP of Engineering, Bruno Oliveira explains how Noema uses generative AI to train computer vision applications in this second installment of Noema’s video series, “Changing the World with Computer Vision."

The follwoing is a transcript of the video:

"The big thing in AI right now is generative AI, meaning any type of AI model that can generate content from audio to text. Actually, it's not a new thing. It's been around since the 60s. There were these simple chatbots that were very limited, but now with the advent of large language models like GPT, and of course ChatGPT as a resulting tool from that, I think that's when AI started to be really mainstream. Tools like ChatGPT becoming mainstream and reaching a broader audience I think we'll open many doors for other AI-based tools to become mainstream as well. I think soon, and it's actually happening already, we have plenty of projects for that in DSR, but advanced chatbots will become a commodity and they will be everywhere. 

Every major store will have them online. There's entertainment, of course, video generation, text-to-video generation, all sorts of content can be now done. Being a very powerful tool, of course, it can be used to cause harm. So when you are able to generate any type of visual or audio content, you can use it to cause harm, you can use it to cause misinformation, which is a big topic nowadays. But I believe that as any powerful tool, it just needs proper regulation. And they are already starting to exist, actually. For instance, in China, deepfakes have to be marked, and in Europe, any social network is forced to remove any deepfakes they can find. The big advantage of these tools for us is the edge cases. 

 For example, the case of flooding, there's only so much footage online that can be used to train our models, and it might not be enough, so if we can generate it, it opens a lot of doors and allows us to improve our model accuracy much faster and much simpler. This applies not only to flooding, but any type of disaster or human-caused harmful situation. It's not that easy to find footage, so with generative AI we can easily do that and help us cover the other corner cases that we wouldn't be able to do so if we only relied on real data. "