RE: https://discuss.systems/@commodore/117147981487761056
This all sounds very amazing but what are the training data? LLMs are good at filling in blanks so that new information matches existing information, they are good at matching and creating new members of an existing set that won't stand out as different.
So you want to keep people docile with an LLM you'd need data on what makes those people docile. You want people to buy a product? You need data on what caused previous purchases.
You want to stop a labor union? You need data.
@futurebird it reminds me of selective pricing, if stores could decide product pricing to perfectly match the maximum amount each individual would pay, they would... but we have rules against selective pricing. We don't have rules against selective information that perfectly curates the exactly what keeps each individual docile or enraged (depending on intent).
