3 ms·
As you can see, I have no overarching theory here; it just seems like a topic worth thinking about. Pulling this quote in because I think it is important. Th
by taeric 5y ago
As you can see, I have no overarching theory here; it just seems like a topic worth thinking about.
Pulling this quote in because I think it is important. The post is a signal amplification of something worth talking about. Not a call to tear down some institute of textbooks.
On the topic called out, though. I'm sympathetic. I recall in grade school a classmate missing the question "without air resistance, what happens to a piece of paper you throw out the window of a moving car?" To me, it was somewhat clear that the "without air resistence" part was important, but my classmate missed it. Others ridiculed them for missing the point, but the more interesting question, to me, was what caused something thrown out to fall behind the car, and how did that relate to the question? That is, on hearing the wrong answer, I was more interested on following up with "and why is that?" Not jumping to "stupid! That is air resistance!"
How do you pull this to interesting topics, though? Programming, in particular, is interesting in that moving to "first principals" is stupid hard for anything graphical. Unless you are wanting to build an engine, you are far more productive using an existing environment. And, basic environments are much harder to force people into today; compared to when the computer booted to a prompt and didn't really store anything for when it was turned off.
Then there is modelling and general ML. Turns out, simple models still dominate all but the most advanced engineering today. And there is no real reason to think that will change. Sure, with enough data, I think deep learning is borderline magic. Turns out, even places that have terabytes of data often have far less relevant data. I'm convinced there is a Nyquist/sampling relation out there for most data. But... I have no idea what that would be. So, we are stuck with simplistic housing price models for the time being. All the while, real estate is basically "whatever sold in the neighborhood recently is what you can expect."
- semi-extrinsic 5y ago> simplistic housing price models I'm pretty sure the next step here will be to build algorithms that pull in colorized 3D data of comparable resolution to what Google Maps has today. Then the algorithm should be able to classify stuff like "this is a nice garden with sunlight all evening", "this has a nice view from the living room", "this one has windows facing the neighbour's windows" etc. I recently pulled down Google Maps data for my neighborhood into Blender (using Chrome+RenderDoc and a plugin for GMaps import). To a human, the resulting colorized 3D model is more than enough to understand most there is to know about the exterior of a property.
- taeric 5y agoI'm pretty sure the largest factor to consider is the buying power of people in the city. Just look at the Seattle market, where houses often goo for +30% of asking. The houses are definitely not magic compared to other similar cities nearby.