Recent advances have been made in Artificial Intelligence (AI) and Machine Learning (ML). We’ve been exposed to all kinds of AI and ML for a long time, even the keyboard on your phone uses ML to guess which key you were trying to type and which word you were trying to type and correct the best it can. The more advanced systems we have now can be leveraged to create engaging content by advertisers (and much more). First, it is important to have an understanding of how these machines work.
We have consumer-facing Large Langauge Models (like GPT 3.5 and now GPT 4.0) which have a huge database of information taken from the internet and have been trained using a method called unsupervised learning on diverse datasets from the internet. During training, the model learns patterns, associations, and structures within the data. It doesn’t have an inherent understanding or awareness; rather, it generates responses based on patterns it learned during training.
The way that the model learns (training) is once the data has been “tagged” (identified) by humans, the model is fed that information, and then as a test, it is fed more data (untagged) and has to correctly identify it. The process is repeated over and over again until the model is correct enough of the time. I encourage you to read this The Verge article and video by CGP Grey on YouTube, as they are helpful in explaining this concept.
Models like ChatGPT and Bing Copilot (both of which use versions of GPT 3.5 and 4.0) have been fed and trained on a massive quantity of data and can generate responses to prompts that users provide. It can generate code snippets, summaries of movies, and much more. It can also be used by marketers. You can ask it to generate a buyer persona for your product, the marketing copy that will be used in ads, or captions for your social media posts. Going further, you can use it to generate ideas for how to frame your product shots, or what the best time to make your posts could be.
It’s important to remember that the responses generated are not always factually correct, and should be independently verified before posting. It appears Humane, the company behind the AI Pin, may not have verified some of the facts that were shown in their release video. The pin was not able to correctly provide the number of calories in a handful of almonds or the best location to watch the upcoming April 2024 eclipse. The company has said it will release an updated promotional video with correct information. This was not the greatest of introductions and certainly underscores the need to verify information obtained from AI models.
Sources:
https://www.theverge.com/2023/3/26/23655456/chatgpt-bard-bing-ai-race-text-boxes