Teaching people how to use AI for maximum benefit

Knowledge is also our shield

A question here yesterday about qualms about a Pxy response to a complex set of queries elicited the response I post below. One worry was about Pxy's response to the role that China played in the American opioid (fentanyl) crisis and excess deaths, presumably opioid-caused.

I post this to explain why I post so much here about using Pxy and how to get the best out of Pxy and AI generally.

1.Pxy can be asked any question, including any hypothetical question. Pages of text (a few, dozens or hundreds), essay or analysis can be pasted in and questions asked about that. A Word (maybe also pdf) file can be attached to a set of questions. The vast range and structure of questions that AI can be asked is a big part of what I am desperately trying to teach people. That is why I keep showing so many Qs&As. It's not me being pompous or arrogant. It's me desperately trying to get people up to speed.** The thought was that if people see how it is done enough times, they will be less intimidated about trying AI for themselves. 

** The vast amount and sophistication of political demagoguery, lies, slanders, crackpottery, irrational emotional manipulation, etc., in the MSM and everywhere else is now overwhelming. Old-fashioned Google searching, now fully crapified for profit, doesn't come close to what is needed as a defensive response. As best I can tell, people's best defense is knowing how to use AI in self-defense. AI is probably the average person's best defense against the dark arts in politics, and probably everything else too.

It's better to teach a person how to fish than it is to give them a fish. Hence, I talk a lot about Pxy and AI generally.  

2. AI query sets must have a detailed error and bias reduction protocol added to the end of each query set. The protocol I coaxed out of Pxy is posted at this link -- bookmark it and use it every time: https://dispol.blogspot.com/2025/08/the-final-ai-error-reduction.html. Add that instruction set to the end of each new set of queries, otherwise the response tends to seek out your own bias and feed it. Pasting in the accuracy protocol once per thread works for Pxy, i.e., the first set of questions gets the accuracy protocol, and that is automatically applied to subsequent queries in the same thread.

Regarding the accuracy protocol, I have repeatedly fed it to Pxy for reanalysis and critical revisions. I also show Pxy its own errors and ask for whether the flaw that gave rise to a particular error rises to the level of needing to be put in the accuracy protocol. I am now at the point of diminishing returns and don't make many new revisions because Pxy says it would be counterproductive. "Asymptotically impossible" is the way Pxy describes further protocol attempts at perfection.  

3. Do not use pro search mode for politics questions. Most of Pro mode answers are too shallow and too error or bias riddled. For politics queries, always search in research mode with the "Academic" and "Social" data sets included with the default "Web" data. Academic get you scholarship and Social sometimes gets you some needed human nuance. Politics is too complex for Pro mode (1) searching, and (2) analysis. Searching and analysis are two separate things, with two possible separate query instructions. The accuracy protocol is my default analysis guidance, and sometimes other factors are separately added for better focused AI analysis. 

4. By now I know how to ferret out an AI response that is more credible regarding China and fentanyl. Maybe the response would be significantly different, but maybe not. That answer has already been passed through the accuracy protocol. Coming up with a better way to ask the question is the kind of learning one has to take time to get a feel for. Learning to ask questions correctly to get the most likely mostly right answer takes a lot of time and practice. Well, at least for me it did. It would take significant time to dive deeper into China and fentanyl related to housing. It would probably require a series of complex query sets, each of which gets run through the accuracy protocol.