Published on November 10, 2025 3:47 PM GMT

I woke up, lights still on. A quick glance at my phone told me it was slightly past five in the morning. So I did manage to sleep, but not much. Probably worse than nothing. The coin was glowing blue again! Maybe I hadn’t broken it after all. It didn’t levitate anymore, though, but perhaps it was still showing the old result.

The experiment felt a bit silly now. If it really took hours between attempts, it might take me a few days to get 99% probability. But then again, if the coin really was magical, I couldn't be sure that results from a simple experiment like this transferred to other domains at all. Still, even predicting a modern random number generator based on a high-level description of what was going to happen was already impossible. In the improbable case I wasn't going crazy, this would net me a Nobel or two.

I was a bit too impatient to actually go through with more repetitions right now. Maybe I could ask some meta-questions. You know, like asking the genie for more wishes. Nothing wrong about that, surely. I grabbed a used envelope and scribbled:

The coin should land tree side up if I would end up believing all of the following statements after testing different ideas over 1000 coin tosses. Otherwise it should land on the other side. "When the coin glows blue, it's active and can be used to get an answer. The blue glow disappears when an answer has been given, and returns after the coin is ready to answer again. The answers given by the coin are always correct."

Hopefully it worked on hypothetical questions too. The formulation felt clumsy and lawyerish at the same time. I might still believe false things after 1000 trials, yet I couldn't figure out how to formulate the statement using some objective benchmark. I could of course add more people, better suited to tasks like this than me. But there was some elegance in using my own understanding for this. Mostly because I wasn't sure I would be showing the coin to anyone else, soon or maybe ever.

I picked up the coin, stumbled my way through the wording, and tossed. Flash, fizzle, a soft "cling" as the coin landed. The tree side was up. Either I was making good progress on figuring this out, or I was too competent in misleading myself. The thought of objective measurements led me to set a timer on my phone. If I figured out the recharge time, I could predict how quickly I could make progress. And maybe asking it if the process could be sped up, although it would still be quite hard to figure out how to do that, even if I knew it was possible. But meta-improvements were always the gateway to scaling.

Both times the coin had landed so that the image of the tree was up, though. What if it just always landed on that side, and the magic wasn't anything that complex? I was about to pick up the coin to check that, when I realized that moving it might change the recharge time. Was I willing to disturb my first experiment to do this? Why not? It's not like I couldn't measure the time later and compare, anyway.

A couple dozen tosses of the now ordinary-looking coin invalidated that line of thinking. It seemed to land equally often on both sides, although the streaks were suspiciously long. But that was to be expected; it was just bias.

There was nothing to do but wait, and make some breakfast. Thinking of the next question made my head hurt.