ChatGPT has been dominating my news feed for the past couple of weeks, and it's truly inspiring. It's able to describe what simple source code does, and to mimic the style of various writers, like Shakespeare, and styles, like a limerick or a haiku.

Surprisingly, folks are keen to ask it factual questions, and find that ChatGPT is perfectly happy to confidently lie in its answers, and many are devoting time to figure out how to remove various -isms from its responses (racism, sexism, etc.)

When I think about where this technology can really shine, though, I see massive potential in interactive art. That is, video games! It's not practical today, but if we get to the place that consumer video game systems have 24-32GB VRAM (on top of whatever is needed to render the game itself), it would be possible to run models like ChatGPT in a limited video game context. Coupled with technology like Whisper, it would enable an entirely new role-playing experience, with models like Stable Diffusion generating textures for environments, whisper translating the player's speech to text, and feeding that to ChatGPT-like models that formulate responses as various characters in the game, as well as a responses from a game master. Speech synthesis models are good enough to synthesize various in-game voices, completing the "dialogue with a computer" experience.

I think entertainment is an ideal stepping stone for AI because it's an application that is relatively low-stakes: not every detail needs to be precisely factual or correct to have value...so long as the world is interesting and immersive for the players, AI will succeed.