[aparsons]

I hate to use this language on HN - but this post is a crock of shit. This is just an opportunistic post by the author that cleverly evades defining anything in solid terms, providing even one concrete example, and is a pot of handwavey BS. It started off encouragingly, and two paragraphs in, the smoke alarms in my head were blaring. The author doesn’t even have a clue (or for some mysterious reason didn’t mention) about what went wrong for Zillow.

Zillow’s massive screw up was that their key business assumption was a self fulfilling prophecy. If a model learns that “price to pay today is price to pay yesterday in this area * 1.05”, and Zillow is buying multiple homes in a neighborhood spread over time, you can see that the price is headed to the moon (and the buyer who puts their faith in the model to the gutter) in no time. In my parents’ town, they offered a neighbor a price, and came back 3 months later with an offer $70k higher. I have a strong suspicion what was supposed to be a price prediction model ended up being a price discovery agent due to the lack of a “visited” set. Oops!

[version_five]

I agree with your characterization. I've been in meetings with people like this that will throw out irrelevant stuff (topology, distributions of distributions!?) as if it's an insight, and there is a certain kind of "leader" that loves that stuff, to the frustration of everyone who actually has some background in the area. I assume this is where he's coming from