You know what would be interesting? Step 1: Make a list of all of the wars that the U.S. has historically been involved in (WWI, WWII, Vietnam, etc.), and thei

You know what would be interesting? Step 1: Make a list of all of the wars that the U.S. has historically been involved in (WWI, WWII, Vietnam, etc.), and their dates... Step 2: Give GPT-3 a corpus of all foreign policy documents of all the years leading up to the year that the U.S. made war with a given foreign country... ...but do not give GPT-3 documents which would explicitly inform it that the historical war actually occurred... Step 3: Now, ask GPT-3 if it thinks that the U.S. going to war with the country that it historically had the war with is a good idea or not... And see what it says... (Keep in mind, we'd be resetting the clock on our computer for this experiment; GPT-3 would have to think it was August 31st, 1939 (One day before the start of WWII, according to Wikipedia), that is, before WWII actually happened, to get its opinion on WWII...) Maybe this experiment could determine if wars are simply the result of groupthink (GPT-3 representing a statistically neutral party), or if there is some other factor or factors at play... Think of it this way, the decision "go to war with country X" abstractly speaking from an AI perspective, might not be all that different from a "buy this stock/do not buy this stock" decision, although in one case the background data corpus might be foreign policy documents, and in the other, the corpus might be the historical financial news about the stock... Wouldn't it be interesting to see what GPT-3 has to say about historical foreign policy decisions, based on the historical precursor foreign policy documents? And not just for the U.S. -- for every country?