A bad abstraction would have caused many updates in many places because the API would never quite stabilize due to having been a force-fit from the start.
A uses the abstraction, but finds the API doesn't work. Fixes that.
That causes B to have to make a tracking change which induces a bug. B realizes that the API isn't quite right. Fixes it.
That causes A and C to make tracking changes. These induce more bugs. C fixes the abstraction to avoid these cases.
This breaks A and B so they decline to update.
And so on. This is what a bad abstraction looks like. API "fixes" bouncing around the code as they reflect off of the bad abstraction.
I, on the other hand, have had to burn through countless cycles of security alerts because I used a library for JSON parsing that had all kinds of other features that I didn't need or want.
The security bugs were all in features I never wanted.
A bit of simple duplication would have been golden.
Add in the fact that they claim 900 million weekly uniques. Pretending the growth and cost rates compound as described in the article, they will need to generate about 100x current revenue to growth out of their current hole. That sort of implies that they will have 10x the entire world's population as weekly uniques at that time.
This is an important point. In the 1980's, PV panels extracted 5-10% of the incident solar energy which could be converted to heat at roughly 100% efficiency. Solar thermal collectors collected at 80+% efficiency and could store and return the heat at about that level for a net 70% round trip. That's a lot better than PV, especially if the collector is your entire south-facing facade.
Nowadays, panels are sitting at roughly 20% and heat pumps have a coefficient of performance around 4x. If you need a battery round trip, you are right about the 70% point and you now have electricity which is more generally useful than low grade heat.
Those 40 year old decisions, as you say, have had several decades of ossification, though, so it is hard to uproot them.
The adoption rate in places like Australia and even Texas is what demonstrates that the argument holds water.
People wouldn't be rushing to shift entire markets at the observed rates if the economics were upside down. It is the soundness of the economic model that is driving the adoption even against tariffs and subversion by the current US regime.
It would be pretty easy to put a DuckDB data source into this code.
It might be pretty easy to use overloading to get special case implementations that form SQL queries progressively until the results need to be materialized as something like a dataframe for the function code to work on.
Are you saying that you have not observed these things in the world? I definitely have. The blog didn't do the work for you, but if we look at some of the claims I think it is pretty clear:
a) increased training scale would result in highly fluent systems that would fool users into trusting untrustworthy output.
Can you possibly be claiming that this is not a common experience? Do you really need references to the legal cases which had hallucinated legal theories and citations? Or the utter slop being passed off as research papers?
b) large-scale AI would amplify bias in the source material.
The large investments nearly every frontier model development team spends on this problem is probably good enough evidence. Grok is another point of evidence. The studies showing that AI systems imitate gender bias in evaluating resumes is another. The gender bias in estimating names of people in sentences is another.
The blog actually mentions specific cases that exhibited all of these problems. They did not cite references for them, but you can use a search engine.
c) environment costs
This is widely discussed and documented. Take Xai's use of polluting turbine generators for their data center in for Collossus 2 in Mississippi as just a single example. Do you really need a reference for the environmental impact of the proposed data center in Utah that (as planned) will consume more energy than the entire state currently does?
d) training set audits are impossible.
Do you need substantiation of the inappropriate imagery in training data? The blog gives you a pretty solid reference.
... and so on ...
I suppose that it could be true that when you say "I don't see" you really meant "I didn't look at the blog". Is that why you can't see the substantiation?
It's the physics of cooling the beasts and the communication delays that make those plans ludicrous.
To turn your assertion on its head, the fact that the supporters don't seem to be able (or willing) to do the math to fact check these proposals is not an indicator that the plans will work.
As a starting point for comparison, the total power budget of the ISS is under 100kW and a single supercomputer rack dissipates about 4x that. What changes to the ISS can be made to get 100x more power and dissipate 100x more heat?
https://www.nejm.org/doi/full/10.1056/NEJMoa2516491
There's your hard data.