I was writing a short story about a VR computer game where you can bike around cities with celebrities, and I decided to have GPT-3 generate a description of James Franco. The amount of detail was astonishing! How can GPT-3 be aware of the physical appearance of people? Did the ML have access to photos? I edited some stuff in, but all of the physical description are generated by curie and davinci:
I decided to check him out while he was still in non-active mode. It was not the first time I played, but somehow I still couldn’t come to grips with the reality of it.
He looked at me and followed me with his eyes but he was not able to interact yet. His face revealed no emotion regarding my presence. The avatar somehow radiated all the qualities I had projected on him: expressive, sensitive, vulnerable, complex, curious, arrogant, insightful.
When I stood in front of him, I could hear his breathing. Over his face of classic Hollywood star quality lay a friendly, contemplative but serious expression. I noticed the sharp jawline and a scar on his right cheek from a car accident. His skin was light brown and the sparkle in his hazel eyes hinted at something outlandish and provocative.
I observed the face I knew so well; the slight cleft chin, the prominent brow. He had a small, thin mustache. The ebony haircut was styled in a messy, tousled way, a few strands of hair protruding over his forehead. His lips were curled in a smile both enigmatic and self-aware.