Assorted observations of mine that don't merit an entire page devoted to them.
- Under our current understanding of physics, the modern environment in which humans have kids fairly often and the population grows over time is a very weird and unsustainable state. The number of atoms under control of humans over time is O(t3), and any exponential function will eventually outpace the bounds set by the speed of light. No matter the wonders you imagine in a post-Singularity future, you'll still have to pick between inevitable death and the chance to have children more than once in a billion years. (This makes me pretty sad.)
-
Something like the Dunning-Kruger effect is actually exactly what you expect from well-calibrated individuals, if your measurement of skill is at all noisy. The people who score lowest on your test will have done so out of a combination of poor skill and poor luck, so their true skill will be higher than what you measured - it isn't surprising if their self-estimated skill is too!
- One can isolate decidedly irrational behavior with more careful setups - I'm not claiming people are actually calibrated about this sort of thing - but "the worst performers at a task think themselves better at it" ought to be your default expectation and not a surprising result in and of itself. (Poor performers giving themselves higher expected ratings than mediocre performers isn't explained by this model, though.)
- Betting according to the Kelly criterion does not maximize your expected profit in the limit, it just maximizes the value of your median outcome (or any fixed percentile outside of zero and one hundred). You can do the math here to confirm that the EV-maximizing betting strategy is to always go all-in on every bet with positive expectation. People extol the virtues of Kelly betting a lot, often justly, but I've noticed a pattern where the claim gets exaggerated to "Kelly betting is the best thing no matter your utility function" and this is not true.
- Online poll responses only give you responses from internet users, which introduces some biases; that much is obvious. Something I think gets considered less is that online poll responses are weighted by time spent online, or in particular time spent on the relevant platform: someone who spends 8 hours a day on Twitter is 50x as likely to be included in a poll response compared to someone who spends ten minutes a day on the site. (This also applies to the writers of online content; see this reddit post for another framing of this effect.)
- Two of the most common ways to tell if you're in dreaming - counting your fingers and trying to read text - are also standard ways to determine if an image is AI-generated (as of February 2023). I think this isn't entirely a coincidence, and reflects something about how fundamental something like the system 1 / system 2 distinction is. Spiritually related: humans who are not concentrating are not general intelligences.