Script theory does a good job of showing how being able to follow certain social scripts does not suggest that the person does not hold a strong delusion that is unchallenged by those scripts.
You can believe you can fly while not needing to contest that until it becomes an active choice you are forced to make. Some might back off from the belief when put to contest, because they are okay believing whatever feels good despite the dissonance they feel. Some might take action believing they can fly. Both exist comfortable in a shared social group of online “flying truthers.”
People are easily deluded, and usually those delusions cover vaguely bounded and varying social groups, where a single person and an LLM can confirm the person into a noticeable delusion that hasn’t been socialized into the surrounding social scripts.
This is also how you can get a genuinely smart person in math or something, which also holds confident deluded beliefs on some other context.
On the whole: Eco-niche specialization, misplaced confidence, and social affirmation, generally can lead to this.
Almost like you need to take different empirical observations from varied tools and perspectives to get more reliable predictions about the wider body. Enough robustness gives you confident weightings that can be used to grow more empirical evidence to build new cognitive tools. No map is the territory, so robustness and weighting need to be an active process in changing growing areas of understanding. no new tools are possible without philosophy actively constructing along science using wider Bayesian basins than some single scientific data point. those varied but well-weighted Bayesian networks are not “just philosophy” like joe rogan giving a very shallow, non-robust, greentext level take on something that sounded mildly plausible.