The Pitfalls of Personal Verification | Generated by AI
Question: Yin Wang’s epistemology — “until I can try it myself, I’ll suspect it” — seems valuable. He doubts the 1969 moon landing and suspects SpaceX landings are staged, but he doesn’t doubt the iPhone, and he correctly called out Tesla Autopilot problems in 2018–2019. Is this a good way to think?
Answer:
I’d split this into two things, because one is genuinely great and the other is where the method silently breaks.
The good part: empiricism over authority. “Don’t trust claims you can test but haven’t tested” is exactly the right instinct for an engineer. It’s why Yin Wang’s Autopilot criticism aged well — he could reason about the actual failure modes (vision-only perception, edge cases, the gap between marketing and the control stack), and reality confirmed it: the 2018 Walter Huang fatality, NHTSA investigations, years of “FSD next year” that didn’t materialize. He was evaluating a testable engineering claim using domain knowledge. That’s first-principles thinking working as intended. Same reason he doesn’t call the iPhone fake — it’s in his hand, it works.
Where it breaks: “I haven’t personally tested it” ≠ “it’s probably fake.” That’s not skepticism, it’s solipsism with extra steps. The correct first-principles move for things you can’t personally test is to evaluate the evidence chain and its independence — and by that standard, the moon landing and SpaceX are among the best-verified events in engineering history:
- Apollo left retroreflectors on the lunar surface. Anyone today — including observatories in China and amateur groups — can bounce a laser off them and measure the Earth-Moon distance to centimeters. This is a repeatable experiment you can run now, which is exactly Yin Wang’s own criterion.
- The USSR, whose entire geopolitical incentive was to expose a fake, independently tracked Apollo with their own radio telescopes and conceded the landing. Adversarial verification is the strongest kind.
- India’s Chandrayaan-2 and NASA’s LRO have independently photographed the Apollo landing sites, hardware and rover tracks visible.
- For SpaceX: booster landings are watched live by thousands of independent people with their own cameras at Cape Canaveral; amateur satellite trackers with hobbyist radio gear independently track every launch to orbit; and Starlink terminals work — you can buy one and get internet from satellites that supposedly don’t exist. The “capsule dropped in the ocean” theory fails a trivial physics check too: reentry footage shows plasma heating from ~27,000 km/h orbital velocity, which no aircraft drop can produce.
So the irony is that applying his own method correctly — find an experiment you can run — confirms both. Lunar laser ranging and a Starlink dish are the “try it yourself” for spaceflight.
The failure mode to watch for (in him and in yourself): “verify by hand” degrades into “distrust anything outside my hands.” A working epistemology needs both legs — direct testing where possible, and rigorous evaluation of independent, adversarial evidence where not. Scientists don’t trust the moon landing because NASA said so; they trust it because five independent chains of evidence, including hostile ones, all converge. That convergence is itself data.
Keep the Yin Wang instinct for engineering claims in your domain — it’s why his PL and Tesla takes are sharp. Just notice that outside his domain, the same instinct produces confidently wrong conclusions. That’s a useful lesson in itself: strong reasoning heuristics don’t transfer automatically across domains, and the smartest people are often the most convincing when wrong.