We've been in the AI game for years
You probably didn’t know that Cleo’s been using AI for over 6 years. True story.
When we came up with the idea for Cleo, we thought wouldn’t it’d be great if you had an AI assistant that made the money thing easy.
It could keep track of your bills. Help you pick the best products to use. Even coach you into some new habits. And do it all with some personality 🤌
6 years later, and the whole field of AI has made a big leap forward, to that point that Cleo is closer than ever to what we imagined her to be.
Side note: If you want to read more about the technical side of things our Head of Data Science has got you covered.
How it started
The first thing any AI model has to do is figure out what you’re actually trying say or get done. This is called your intent.
We spent a lot of time helping Cleo's AI use machine learning to figure out your intent and match you with the right chat flow, usually written by a real life comedy writer.
It wasn’t the fastest process, but it was a labor of love.
How’s it going
Then all of a sudden (well it seemed suddenly, but it was decades in the making) large language models, like ChatGPT became real interesting.
These models offered a huge boost in two key places.
- Understanding what our users were trying to say (natural language understanding)
- Creating responses that better fit the context (natural language generation)
So when we took all the work we’d done on Cleo's LLM and sprinkled some new LLMs on top, the results were wild.
Cleo could chat about more things than we could ever write ourselves. And do it in a way that’s super smart and personal.
Behind the scenes
“So when we chat to Cleo we’re just messaging ChatGPT?”
Not quite. There’s a bit more secret sauce in there. They key isn’t just sending your info over to another model, it’s sending the right info at the right time.
The ChatGPTs of the world are best at general knowledge stuff. So the first thing Cleo's AI does is determine if your question is even something they’d be good at answering.
If it is, the next step is to package up the relevant info or point them to the database they need to craft a response.
For example, if you say “I need help saving money” Cleo would share your biggest splurges. Which we all know is Uber Eats.
My data is still safe right?
100%. Cleo is not a snitch. We don’t just send your data out to random AIs. Every model we partner with goes through a vetting process.
And for the really important stuff, Cleo handles most of it herself.
Keeping it Cleo
Not everything goes to ChatGPT. I mean, c’mon, what kind of business model is that?
Whenever you ask Cleo a very specific question that needs a definite answer we’ve already trained our LLM to recognize the intent and give you a solid reply.
These questions aren’t general knowledge and have a definite answer or require knowledge very specific to Cleo.
“What’s my credit score?”
“Can you cover me $100?
“Change my payment method”
There’s also some quirky things about Cleo that people love and that you can’t find anywhere else. Roast mode, anyone?
The future of AI at Cleo
So if this sounds like we’re pretty much getting to have our cake and eat it, you’re kind of right. Taking our last 6 years of AI work and connecting with other LLMs has been next level for Cleo.
But at the same time, she's still herself. Her personality is still there. We still use a lot of the same chat replies that come from the writers room. And we’re even writing more replies to help the other LLMs better understand our voice.
And we're just getting stared. Alll the work we’ve done to understand your money is about to come together in some really big ways.
Watch this space.
*OpenAI is not used in conversations related to the credit builder card, issued by WebBank.