‘Personality goes a long way’. I heard that once in a movie. It’s as true for generative AI as it is for real human people.

While much has been said about ChatGPT over the last six months or so, the reality is that there’s plenty of choice out there when it comes to engaging with conversational AI, and there are notable differences in character and personality between models. I won’t attempt to disambiguate models, ‘fine-tuned’ models, engines etc. The terminology gets pretty esoteric, pretty fast.

Let’s focus on the consumer layer that you interact with when you visit chat.openai.com, bard.google.com or similar. My observation, having tinkered with a few different sites and apps, is that you might not want to pledge allegiance to just one. Rather, you’ll need to choose the right tool for the job. For any given conversation or context, you’ll need to factor in your goals and how you plan to use what the AI tells you. You might even want to consider your emotional needs at the time.

ChatGPT is indeed somewhat chatty, even verbose, but it often makes a point of distancing itself from the notion of personality. For example, given a simple, deliberately broad question (‘what’s the best music festival in the world?’), to which the answer must necessarily be subjective, ChatGPT answered: ‘As an AI, I don't have personal opinions, but there are several music festivals around the world that are renowned for their exceptional performances, unique experiences, and overall atmosphere. The concept of the "best" festival can vary depending on individual preferences and music tastes. Here are some of the highly regarded festivals:’

So it spent over 50 words emphatically distancing itself from having a personal opinion, and to set itself up to provide a list. Its imperative was to provide a result, more than to hold a conversation.

The same question to Bard elicited a response beginning ‘The best music festival in the world is a matter of opinion, but some of the most popular and critically acclaimed festivals include:’

So Bard took fundamentally the same approach as ChatGPT, but delivered its setup in fewer than half the words. It seemed slightly less hung up declaring its objectivity and keen to crack on with the answer.

Pi (heypi.com) took a different approach. Pi prides itself on having a personality; it’s its raison d’être. Therefore, although Pi tries to be objective, it also talks to you like a friend. Pi’s answer began ‘It depends on what type of music you’re into, but there are some pretty amazing festivals all over the world.’ A less explicit disclaimer around personal opinion. This approach makes it seem less clinical than some of the other AI tools. To this end, Pi will also not rush to provide a glut of output, or a long list. It will frequently pause and prompt for your feedback. It understands conversational turn-taking better than the others and seeks to keep up back-and-forth ‘chit-chat’.

You could interpret the answers so far as akin to that of a politician, an advisor and a friend, respectively.

Needless to say, different prompts elicit different tones and expose different capabilities. If you need one of these tools to write some code for you, such as ‘write me some code to validate an email address’, both ChatGPT and Bard will do so. Pi will refuse - ostensibly because it’s a person - and people (even virtual people) don’t just spew out syntactically correct code, character for character, in the course of conversation. It’s not a very human dialogue thing to do.

‘I'm afraid I can't code, and I'm not able to validate email addresses. My expertise is in conversation and understanding language. But I'm sure there are lots of great resources out there for learning how to code.’

…and fair play. The Pi team knows how they want to position their personal AI.

Other tools I have tried have yet different personalities. The process of interacting with Jasper AI, for example, is more like interacting with a knowledgeable, efficient colleague and Grammarly more like interacting with an omnipresent tutor, intentionally so. Different tools, different contexts, different user experiences, different tones. All AI!

When you’ve finished an interaction with ChatGPT, you’re likely to think ‘that was really helpful’. After chatting with Pi, you’re likely to think ‘what a thoroughly nice character’

It’s worth noting that all of these AI tools set out different charters in terms of their ethics, purpose and principles - something to factor into the decision making process.

The point for us here at CV Wallet is that as we build out our career management app, we want to provide tools that will serve up the most helpful and engaging experience. So our AI tool for writing a personal statement might need to be polished and efficient, whereas our career advice chat, more conversational. It’s our job to curate and blend the right models and transformers, and package them into a seamless in-app experience. We’re on it!