Discussion about this post

User's avatar
Andrew Schreiber's avatar

Yes! Now is certainly the moment for the Cambrian explosion of AI characters.

Building atop generative AI introduces a novel dynamism that changes their appeal. The possibility of endless content makes a character potentially infinitely engaging - users will never run out of dialogue. At the same time, generative AI also creates a consistency problem as users push on the edges of the LLM prompt - threatening the suspension of disbelief. It's fun meeting a costumed Micky Mouse at Disneyland, but if you talk to the masked man long enough, you may wind up disrupting the carefully curated image of Micky Mouse that teams of people at Disney are trying to create in your mind.

In the context of values, AI characters will naturally fall back towards the embedded values of the foundation model whereever the prompt is under-descriptive. Some characters will be fixed to a certain model, but users will probably expect the lucidity that comes with the newest foundation models. It will be an interesting challenge to maintain a coherent value system for a character as the underlying foundation model continues to change.

Curious how you think about the values of an AI companion matching their human's over time. Let's say the person reveals they deeply prefer monarchies over democratic rule - would a good companion nudge them towards democracy or shift itself towards honoring the throne?

Expand full comment

No posts