In yesterday’s NFT I had GPT3 whip up an article about a haiku we had written together from the perspective of Carl Sagan.

What we got back felt deeply profound. It genuinely sounded like his voice, but more… the way it parsed the concepts in the haiku is still pinging the back of my brain the next day. We know it's trained on gazillions of text data points, and that picking the most likely best word across that massive data set makes magic happen… but there must be more going on. It has to grok concepts and concept order to do what it did yesterday!! I need to talk to @ramsus about this. She will have insight I’m sure.

The ramifications for the future will keep me up for the rest of the year. It’s already this way, and GPT4 etc will likely make it WAY better;

WE CAN TALK TO THE DEAD Every major figure in history, moreso for those who were prolific writers, will have AI ghosts lingering in the machine now.

When they train these LLMs on all of Facebook and Twitter… you will have an AI ghost too.
Our ancestors are going to be able to talk to us and feel a sense of real connection and wonder.
Kids born in 20 years will only know a world like that. And someday those ghosts will be arguably sentient.

In a way we will have achieved immortality.