3aIT Blog

A mobile phone with the ChatGPT homepage open in a browserIt is certainly the case that unless you're the sort of person that keeps on top of tech news, that initial explosion in awareness of something significant having changed when the world saw the very rapid progress that had been made in the generative AI world with tools like ChatGPT has now quietened down. It would be easy enough to assume that it was a lot of fuss about nothing. However, scratch the surface and it's clear that isn't the case. We therefore decided to spend a bit of time letting our development team play with the potential of these tools on our own data to find out what they're capable of right now.

Like most people, we played with ChatGPT a bit during the inital surge of interest. It was clearly amazing from a technological standpoint. However, it was a bit like seeing a fundamental building block of something bigger rather than the finished article. Sure, being able to ask it about recipes or history or anything else you care to think about is clever, but other than the conversational style of it, it's not a million miles away from just using a search engine.

Use your data, and everything changes

As you might expect, our primary set of data is a database of tickets. For every project or task we get from a client, we create a ticket and log everything we do. This means we have a vast database of every problem and solution we've ever encountered going back twenty years. We have reports that can easily tell us how long we've spent working for a particular client or how many hours a staff member has logged, but unlocking the potential of the vast database of knoweldge there has always been pretty crude. You can search on keywords and find every time someone has mentioned that thing, but unless you're very specific or looking for something very niche, that can return a lot of noise, and what you want is buried somewhere in all of that.

Someone interacting with ChatGPT on a mobileOnce we'd worked out how to get ChatGPT interacting with our database, it instantly became obvious that we were looking at the beginnings of a fundamental shift in how we are likely to build systems in the future. Playing with a "generalist" GPT tool is impressive, but it becomes something else entirely when it's pointed at your own data. Applied at a database-wide level, it's then possible to "talk" to your database. It can answer the "report style" questions like "What did X do today" or "How long did we spend doing X". However, it can also respond to queries like "Have we solved a problem like this before" and it can suggest a solution based on your knowledge base, or from its own general knowledge if it looks like the answer is no.

Now, before we get too carried away here, it's worth saying that our investigations are very much ongoing here. It does currently need a lot of tuning to provide helpful answers, and that involves a lot of testing and then tweaking its "rules" when it's answering in the wrong way. This is further complicated by the fact that you can ask an identical question twice and get different answers. Hopefully not on the "facts" based questions ("How much time was spent doing X"), but certainly on the questions that analyse lots of text. This can be mitigated by further tuning, but we add it here as a note to anyone that has read this and is getting excited!

What does this mean for your business?

A laptop displaying some generic data on the screenIn the hands of the right person that understands the range of the data a company stores, the potential here is obvious. However, that isn't everyone in a company, or even most people. For these people, the applications will need to be more targetted. Specific fields that can pull in data or suggest text based on a much more constrained set of rules with little to no margin for error. "Write a description for this thing based on this other very similar thing we sell", for example. 

This gets to the nub of where things are now in terms of the business-specific applications of AI. There isn't a magic "AI" button that can be pressed and everything sorts itself. We have the foundational building block, so the question becomes "How can this be applied to your database?" We have a very text-heavy database, so pointing it at that is the obvious route to take. However, that won't be the case for every business. So it will be a case of working with someone that understands your data and can suggest ways these tools could help. If you have a very data-heavy database (almost entirely full of costs, times, dates etc), it may be it has limited application for now versus running a report you can be certain is giving you the correct data first time, every time. However, even that may change in the future. This revolution isn't going to happen overnight, but there's no doubt we will all be using tools like this in the next few years.