I took a different approach using AI. I conducted the interviews and survey, placed all the gathered data in a NotebookLM folder with a detailed project description, and built the personas around that information. What do you guys think about this approach?
I am doing a lot of work with AI personas representing human perspectives for my company Persimi (https://persimi.com).
I have seen many of the above issues first-hand, and it does represent a challenge.
However, we discovered that a significant factor is that as the LLM acts its part, the inauthentic results come from a prompt that does not specify the authenticity of the human moment. We don't tell it about it's aches and pains, or how tired it is, or that money is tight... etc.
In short, the limitations of the prompt have us framing a character on a sitcom (where details and texture are limited) instead of conveying the richer circumstance of a great novel.
Persimi introduced a number of augmentations on our personas, including mood, context, time of day, and others, and saw amazing results.
While we will NEVER say that AI personas can replace feedback from real human being, they are a very promising channel for augmenting that data.
Check out a data set where we change the mood and context and ask the same simple question multiple times. You get a range of answers (or facets to the same answer) that are fascinating.
George Johnson | 120 years Old | What is your greatest fear?
I took a different approach using AI. I conducted the interviews and survey, placed all the gathered data in a NotebookLM folder with a detailed project description, and built the personas around that information. What do you guys think about this approach?
This is exactly why I don’t create personas the traditional way anymore
I am doing a lot of work with AI personas representing human perspectives for my company Persimi (https://persimi.com).
I have seen many of the above issues first-hand, and it does represent a challenge.
However, we discovered that a significant factor is that as the LLM acts its part, the inauthentic results come from a prompt that does not specify the authenticity of the human moment. We don't tell it about it's aches and pains, or how tired it is, or that money is tight... etc.
In short, the limitations of the prompt have us framing a character on a sitcom (where details and texture are limited) instead of conveying the richer circumstance of a great novel.
Persimi introduced a number of augmentations on our personas, including mood, context, time of day, and others, and saw amazing results.
While we will NEVER say that AI personas can replace feedback from real human being, they are a very promising channel for augmenting that data.
Check out a data set where we change the mood and context and ask the same simple question multiple times. You get a range of answers (or facets to the same answer) that are fascinating.
George Johnson | 120 years Old | What is your greatest fear?
https://app.persimi.com/public/reports/?id=87ab2f854288a74fa1d21db066a3a9eb
Great article! 🤗 The way I see it, AI is great for understanding real users in real time (users, not personas!).
It actually makes traditional, static personas outdated and irrelevant. Thought you might enjoy this post: https://open.substack.com/pub/karozieminski/p/user-personas-are-dead-ai-powered?r=gngtc&utm_campaign=post&utm_medium=web&showWelcomeOnShare=false