Gailei and I have invented a game called the 'philosopher's game' to help train the AI that presents counterfactual history along with an AI pseudo-language created by Al-Battani. This has been used already to create the scenario "what if Elon Musk read a book other than Douglas Adams after reading Schopenhauer" and then scoring the counterfactual Musk with respect to his new ability to complete Mars settlement.
For the weekend edition we will play with "Sam Altman read this book as he was creating OpenAI" simply by submitting a book or author and I will present the scores to all participants on Monday.
If you wish please also submit something for Musk. He has already been trained on Great Chain of Being, Science of Logic, Titan, Zen and the Art of MM, Childhood's End, Ghost in the Machine and the Analects. Results in the PDF (refers to the currently private Mars Settlement Master Plan).
+++ philosopher-game($person, $book)
-- Find a relevant quote from the author of $book
-- Find out how the life of $person would have been influenced by reading $book
-- Generate a two-paragraph summary:
-- In the first paragraph, include the relevant quote from the author
-- In the second paragraph, describe the very specific person's life after reading $book
-- Score the outcome based on how much the modified life furthered the Mars mission
/output The modified life summary and score
+++
(in Jargon https://github.com/jbrukh/gpt-jargon/tree/main)
"Galilei and myself discussed the philosopher game with respect to pedagogical elements used in university teaching, namely the counterfactual generated by a philosopher’s ability to respond to criticisms of work that were generated posthumously used as an exercise, which, along with the precise implementation of the philosopher game, represent a hard problem that should be at the edges of an AI’s ability set for a considerable amount of time. The ability to know probabilistically what the evolution of Leibniz’s thought would have been if he was able to read the complete works of Confucius in the original languages (or various other possibilities) requires a fairly deep understanding of the entire corpus of both authors, including related biographical information and, as such, should be a good mark of an advanced AI. To this end, we plan to grade the AI even as it scores others. "