Beyond Chatbots:
The Assured Self Brings AI Personalities to Life
This isn't just another chatbot. The Assured Self crafts AI companions that understand you.
Artificial Intelligence is just a catch all term for any attempt to simulate intelligence. Within that there’s a subset called machine learning and within that there’s what we call large language models which are about the understanding and prediction of text. Which leads into generative AI tools.
When you use a generative AI chat tool, for example Claude, Gemini or ChatGPT, what you’re actually doing is using those companies’ user-interface which is pointed at their specific large language model. The model itself is augmented by what they’ve built around it, memory, plugins etc, things that aren’t part of the large language model.
Because the large language models are very broad, they tend to be great at generalisation, and what you can actually do with your own knowledge and skills is to connect directly to the large language model and train it through a process called “fine-tuning”.
Fine tuning lets you build what’s called a “synthetic identity”. So, a good example is that in the next couple of years you will go to the Disney website, and you’ll have a conversation with Mickey Mouse. That will be a fundamentally different interaction and experience than when you go to the KFC website and talk to Colonel Sanders. And that will come from their back story, their branding, motivation, quirks and whatever else they want built into these characters.
Ultimately everything will have some kind of large language model or chat interface. So when your washing machine breaks down you will talk to it. When you’re sitting next to your Christmas tree, you will talk to it and depending on what the developers have created, it will have its own personality and its own knowledge about itself. This is the sort of thing that we’re going to start seeing in the next couple of years.
If you think of taking a stock response from a large language model, but want to build a therapy app, the response from a CBT therapist would be very different from Jungian therapist. So, we take inputs that would be given to a large language model and then we show the kind of response we want. So, if you imagine that you’re building an AI system, do you want to offer multiple solutions, do you want it to stick to a set process, should it offer feedback?
If it’s an accounts payable AI, are you looking for it to build rapport with your suppliers, or does it just carry out the task at hand, maybe it’s a fan of Monty Python?
All these design and character decisions not only influence the type of AI it becomes, but also influence how people will respond to it.
The final example would be, if you’re building something to train people, is it asking questions, is it prompting discussions, is it just giving people factual information, does it draw the distinction between a seminar or lecture? These are all these things you want to adjust in the base model so that it always acts in a consistent manner.