On 27th November, Scrub hosted an event at Lloyd’s Lab to talk about AI; not just the exciting stuff everyone’s heard about, but the practical side of how it’s used in real life. Generative AI (GenAI) is the buzzword in this space right now, but we wanted to explore how it works alongside non-Generative AI, a quieter but equally important part of the picture.

At Scrub, we’ve spent four years using AI to help insurers clean and validate their data. This event was our chance to share what we’ve learned, clear up some common misunderstandings, and start a deeper conversation about how to get the best out of these technologies.

Here are the highlights from the evening.

Breaking down the layers of AI

AI is not monolithic. It’s more like a collection of tools, each designed for different kinds of jobs.

GenAI is one of those tools. It’s great at creating new content, like text, images, or music, by learning patterns from massive datasets. This makes it perfect for creative or open-ended tasks, like writing a poem or summarising a complex document.

But not every problem needs that level of creativity. Non-Generative AI focuses on tasks with clear rules and predictable outcomes. It’s reliable, consistent, and efficient, just what you need for things like sorting data or making accurate predictions.

Both have their strengths, but the key is knowing when to use each.

What we learned about GenAI and non-GenAI

During the event, we took a closer look at how both Generative AI and non-Generative AI work in different scenarios.

GenAI is flexible and can adapt to all kinds of tasks. It’s especially useful when you don’t have a lot of training data or need to tackle something really complex. But it’s not perfect. GenAI can get things wrong in surprising ways known as “hallucinations.” It can also be expensive to run and tricky to use for tasks that demand absolute precision.

Non-Generative AI, on the other hand, is dependable. It’s been a core part of our platform at Scrub, helping us process messy, semi-structured data quickly and accurately. It’s not built for creativity, but that’s not what we rely on it for. It’s about getting the job done efficiently.

The Scrub philosophy

At Scrub, we don’t start with the flashiest tool, we start with the problem. We break it down into smaller steps and use the simplest solution that works for each step. For straightforward tasks, that might be a rules-based system or a traditional machine learning model. For more complex issues, we’ll explore using tools like GenAI, but only if it’s the best fit.

It’s not about jumping on trends; it’s about solving problems in the smartest, most effective way. That’s how we’ve been able to save our clients time and money.

Looking Ahead

For us, the future of AI isn’t just about seeing what the raw capabilities of the next LLM are; it’s about using it in the most effective way for us. One of the areas we’re excited about is knowledge distillation. It’s a way to take everything a big AI model knows and teach it to a smaller, faster one. That means you get the same results without the big costs or complications.

Thank you to everyone who joined us for this event. It was a great start to what we hope will be an ongoing conversation about how AI can really make a difference in our industry. We’re planning more events like this, so stay tuned for what’s next!

For those who couldn’t join us, we’ve put together a downloadable version of the presentation, covering all the insights from the event. You can access it below.