Leveraging ChatGPT to Find Suitable Content at Scale

Since its November release, there has been a ton of buzz about ChatGPT. There are some who say that it will change the world and many who think it is certainly going to change a variety of businesses by automating many jobs. 

At Pixability, we leverage this AI to help us solve complex problems on fast moving platforms, such as YouTube and ConnectedTV. Every minute, there is another 500 hours of content added to YouTube. There are billions of videos and it would take millions of employees to keep on top of – truly an impossible task. As a result, we rely on a combination of humans and technology to curate brand suitable and contextually relevant content that our advertisers can rely on to reach their target audience. With the release of ChatGPT, this technology has improved tremendously and has had a positive impact on our ability to provide suitable content at scale. 

How Pixability is using ChatGPT to help drive better results for our customers 

At Pixability, we use a combination of our unique insights, proprietary tech, and best in class expertise to help our customers find the content that is most suitable for their brand, delivers on their specific campaign objective and helps them find their key audience. To accomplish this our data science team is constantly managing and updating our data models to improve our ability to identify content. 

OpenAI, the company behind ChatGPT,  has trained their models on a big portion of the internet and other sources, so it is very good at understanding trends, edge cases, and 100’s of millions of different things. For our team who works on large video platforms, such as YouTube, ChatGPT provides a way for our models to ‘learn’ much faster than traditional model development. Let’s use a specific YouTube example. If, in the traditional model, we have trained our models on 10,000 food & beverage videos – but 0 of those videos contain any Mukbang videos. We will not be able to know what a Mukbang video is, or how to categorize it. 

If, in the traditional model, we have trained our models on 10,000 food & beverage videos – but 0 of those videos contain any Mukbang videos – then we will not be able to know what a Mukbang video is, or how to categorize it. (Mukbang is a category of video content where people eat a massive amount of food). 

Today Pixability is using the ChatGPT API to expand its training data to improve our machine learning models. This allows us to create new models quickly and for specific customer requests, without having to wait for human curated training data. One example of this is expanding out contextual category targeting to over 1,000 different categories. In the past, this would have taken 100’s of examples in each category curated by a human. Now, we can quickly create training data, deploy a model, and then tune and improve that model using human inputs. It is much faster and more accurate. 

By leveraging ChatGPT, we can find new trends quickly, and our models can find trends that we haven’t even identified yet. Old ways of creating models would mean we would have to know that PickleBall is a sport, then go out and find examples of pickleball and have a human decide, pickleball, tennis, ping pong on a set of videos. Now, models can understand (without human input) that pickleball is a new sport that is similar to tennis and ping pong, but be smart enough to know that it is different.  

Curious to learn more? Reach out today to discuss how we can help you find the most suitable content for your brand.