[examancer]

You try 3D printing a house on a Pentium, or running inference models. We needed the compute. We need so much compute it boggles the mind and we don't know what to do with it.

Only then, as we are drowning in compute, does it spill over into other areas and allow that compute to be used as leverage towards enhancing or automating areas compute has yet to break into. Only then are crazy ideas like having large clusters of transistors act as neurons in a neural net actually possible, and efficient enough.

This is at least what I mean when I occasionally say something flippant like "(technology/computers/software) will eat the world". Maybe our relentless pursuit of more compute isn't the best way to solve complex problems, but it increasingly seems like it may be a way.

[choonway]

Whether you print a benchy using a tabletop 3d printer using PLA, or a using a gigantic gantry crane pouring cement, the compute power required is the same.

3D printing firmware and all of it's associated design software (CAD/CAM) would run fine on computing technology 10 years old.

Construction speed is bottlenecked by the lack of investment in new techniques, not for the lack of compute power.

[michelpp]

Real buildings must take into account material costs, weight bearing, soil conditions, thermal cycling through various paths, and human and seismic induced dynamic loads. The larger the buildings get, the exponentially more compute is needed to solve these problems.

“Anyone can build a bridge that stands, but it takes an engineer to build a bridge that barely stands.”

[choonway]

You can still design 99% of the bridges in existence today with commercially available software and hardware from 10 years ago.

Why do you want to make a bridge that barely stands?

Instead you want your engineer to:

1. build it cheaply and quickly,

2. maintenance is cheap and non disruptive.

3. probability of unplanned failure is small, eg once in 100,000 years

[fennecfoxen]

Investment in new techniques is limited due to the limited prospects of saving money. Construction is an efficient business; gains are possible but hard, and material transportation prices play a much larger role than one might be used to.

[pbourke]

> Construction is an efficient business

It doesn’t feel very efficient, especially residential construction. My neighbor is building a garage with some finished space above. It’s sat without a garage door for literally months now - everything else is done.

I often wonder how soon into the project was the garage door ordered and how was that order tracked? Was the lead time calculated and a fallback option presented? These are things that we just take as a matter of course in software. They are also routine in industries that are software-intensive (finance, insurance, retail, to a lesser extent logistics)

The construction business - especially residential construction and remodeling - seem insanely inefficient to me. Jobs that should take a few days of wall clock time often take weeks due to poor planning and scheduling.

It’s interesting that certain high value specialties within construction end up being extremely efficient. You can get a new HVAC system (assuming that you don’t need to install ductwork) in a couple of days. Ditto for a new roof or a new driveway. The process is finely tuned.

It would be wonderful to have the ability to apply computational power to jobs that are not as repeatable today. Scan the house and identify constraints. Automatically design the systems, measurements and layout and assemble a BOM to minimize the effects of supplier lead times. Submit detailed RFPs to subcontractors and track their performance to plan. Etc.

[fennecfoxen]

> It doesn’t feel very efficient, especially residential construction. My neighbor is building a garage with some finished space above. It’s sat without a garage door for literally months now - everything else is done.

If all you have to improve the construction industry is better scheduling and supply-chain management, that's something. But the poster above seemed to be hinting at things more like new construction techniques.

[choonway]

haha. the exact same language was used when major semicon foundries was stuck at 14nm for a while.

sometimes solving multiple hard problems simultaneously is easier and yields better results than trying to solve individual problems on their own, due to synergies that can be taken advantage of.

however, the “divide and conquer” adage has stuck too hard into the minds of ordinary people and going against it - that’s the really hard part.