I was in distribution for 400 days straight. Then I skipped.

annika_iyd 9 hours ago

IYD founder here. First, congratulations on 400 days, that's a top-1% streak. Second, I want to gently push back on the framing. The unit didn't judge you. It has no concept of skipping. It only knows that across 1.4 million daily observations, nobody skips. You're not being punished. You're being informed.

four_hundred_green 8 hours ago

I was happy. I got good news and I was happy.

annika_iyd 8 hours ago

And that's valid! The model doesn't have a prior for it in that location, is all. If more people were happy in the quad it would be green. In a sense this is feedback for the quad.

bayes_or_bust 8 hours ago

Statistically, the interesting thing is that the distribution is no longer independent of the measurement. Once people can see the light, they regress to the mean on purpose, the variance collapses, and the threshold tightens with it. Run that forward and eventually blinking is an outlier.

annika_iyd 8 hours ago

We've observed this and we call it "convergence." Customers love it. Incident reports approach zero.

bayes_or_bust 7 hours ago

Incidents approach zero because behavior approaches zero.

annika_iyd 7 hours ago

We should talk. We're hiring.

parking_structure_b 7 hours ago

Level 3 here. We have a book club now. Someone brought a guitar. It's the only place on campus anyone laughs out loud. If you're reading this from the quad: walk normally to the east stairwell. Normally.

grumpy_bsd 6 hours ago

Bentham designed the panopticon so the prisoners couldn't see the guard. It took a YC company to figure out you could skip the guard and just add RGB.

ml_eng_throwaway 5 hours ago

Worked on a similar system. The thing that made me quit: there was a guy who went red every single day at 3pm in a train station. Stood perfectly still for a minute. We flagged him for weeks. Turned out his wife used to meet him there. Standing still is our most common trigger. I think about that a lot.