AI Commanding AI
I asked GPT3 to describe a modern room:
describe in great detail a beautiful modern designed room in 100 words or less
This was the response:
This modern room features a sleek black leather couch and matching armchair, set against a backdrop of crisp white walls. The floor is made of light oak hardwood, and a large, contemporary area rug ties the whole look together. A white lacquer coffee table sits in the middle of the room, surrounded by a pair of retro-style lounge chairs. In the corner, an abstract metal sculpture adds an artistic touch to the decor. A large window allows natural light to fill the space, while a bright chandelier hangs from the ceiling to provide a soft, ambient glow.
It sounded good.
I then took that output and used it in Midjourney to get a few concepts:
I don’t know how to credit the above image. Normally, I put “Generated by author using Midjourney” but in this case, all I did was copy and paste the prompt generated by GPT3.
Now, Midjourney wasn’t perfect at the task. It missed things like the chandelier. However, it was pretty good.
The application of these two technologies could be the rapid creation of simulated worlds. No need to imagine all the components of a virtual room and meticulously model them — just as one AI what should go in the room and then another AI to create it.
This may result in the scaling of complexity in today’s video games and the possibility of video games with no pre-determined outcome.
