AI has been around for decades, but it’s only recently that we are seeing its full potential being realised. The complexities of AI used to be a technical challenge: now they're a product innovation problem.
GPT is not just a tool for implementing clever chatbots or summarising your content. The hype surrounding ChatGPT's ability to create poems, craft ok blog posts, or whip up recipes is merely the tip of the iceberg.
Instead, consider GPT as a tool to add generic intelligence to your existing products and digital offerings.
Here are a few examples of how companies are already using GPT to improve their products and services.
A lot of businesses don’t yet see AI as something that fits within their product, but as something that challenges it. At Planes, we see AI as being table stakes for most digital products, as users expect products to do more work for them and act smarter.
For the techie-readers out there; keep in mind that fundamentally GPT is an API (model) with a string
as an input (the prompt), and returning a string as an output. It’s the input and output formatting that is key to tailoring it to your application.
Online communities and forums are handy platforms for companies like Stripe to enable their communities to help each other solve problems without employing costly support teams. However, where you find users needing help, you’ll likely find bad actors looking to take advantage.
Stripe has adopted the use of GPT-4 for fraud detection in their online communities. It can achieve surprisingly good results when asked to scan and identify patterns in forum conversations. When GPT-4 spots a potential issue, a human can step in and asses whether there is a problem.
Let's imagine a prompt for this:
You are a fraud-detection bot scanning the forums of an online payments provider. You have access to a full thread with the title “<insert name of conversation here>” and need to flag any bad actors.
Here is the thread:
{insert conversation history here as formatted JSON}
You must respond with an array of any message ID’s that you think could be trying to gain account or personal data from another member.
Your output should look like this:
[”message_id_213”, “message_id_799”]
Many online businesses have a tonne of customer reviews, but no easy way of understanding the sentiment or insights hidden within. Yabble, a market research company, used GPT-3 to empower its customers to achieve just this. Using Yabble’s suite of tools, online businesses can ask questions about their reviews, i.e. “Does our product ‘X’ ever break whilst doing ‘Y’ with it?”. Without GPT it's unlikely they would have been able to achieve this without a human somehow memorising every review.
To operate on large data sets and be able to query them I recommend looking into OpenAI’s Embeddings feature.
Planes has been working with an online retailer to retrospectively write Alt Descriptions for all the product images on their site. Alt Descriptions are captions that are read aloud by screen readers, often to visually impaired visitors.
We developed a method that combines an image recognition AI model with GPT. We are excited to implement generative AI in a way that improves people's online experience.
Check out ALT AI now.
For the first time, you no longer need a PhD in Machine Learning to apply AI to your business.
Most product teams likely fall into one of two categories: sceptical or scrambling. Sceptical if you think GPT has no place in your product, or scrambling if you are figuring out how to apply the technology.
Either way, I encourage experimentation at this point; have a play, run an internal Hackathon, and you’ll be surprised by the interesting use cases that spill out.
Planes are running free half-day workshops for your business to see how generative AI can improve your product. Register your interest or drop me an email, henry@planes.agency to find out more.