Chris Said, a data scientist, writes about statistics, A/B testing, scientific misconduct, the Soviet economy, and the air quality index.
Commenters repeatedly point to his Soviet optimization post as a good read on the Soviet economy and cite his air quality index piece.
Parent comment refers Double Descent in Human Learning [1, 2] which references the Deep Double Descent phenomenon in Large models [3].
The gist of double descent is the phenomenon where a _large_ model appears to over-fit as we expected from traditional ML, but then the val loss starts decreasing. The general consensus is that the model switches from memorization mode to interpolation mode which enables generalization.
In essence, the large number of parameters initially fit directly to the data, but then smooth out.
The reason we can continue improving the model is the fact that over-parameterized models always have a descent direction, simply due to the dimensionality of the model.
The paper that the post references, shows that by trading parameters with more training steps, we can have small models that exhibit similar if not identical behaviour.
[1] https://news.ycombinator.com/item?id=35683754
[2] https://chris-said.io/2023/04/21/double-descent-in-human-learning
[3] https://openai.com/research/deep-double-descent
There is an open debate about the likely damage done by power surges along long-distance power lines. Industry funded research claims that they should survive both EMP and major solar storms. Real world experience suggests that they are overconfident.
See https://chris-said.io/2020/06/18/everything-ive-learned-about-solar-storm-risk-and-emp-attacks/
Regardless of which side you believe, it is clear that any "worst case scenario" should include the possibility of an EMP attack that actually does take out the power grid.
I have recalled related post[1] and relevant NH discussion[2] from 2020. Also just ran across another recent related article [3] and its Reddit discussion [4].
[1] https://chris-said.io/2020/06/18/everything-ive-learned-about-solar-storm-risk-and-emp-attacks
[2] https://news.ycombinator.com/item?id=23580362
[3] https://www.sciencenews.org/article/sun-solar-storms-earth-havoc-space-weather-forecasts
[4] https://www.reddit.com/r/space/comments/lt7tnd/solar_storms_can_wreak_havoc_we_need_better_space
This is historical revisionism. The soviet leaders tried to copy the success story of Deng Xiaoping's transitions from socialism to capitalism in China. The planned economy system just didn't work. The "US Empire" can't explain why by the late 1980s, the Soviet Union had 4x tractors as the United States which was totally wasteful.
Here is a good article on why it's computationally challenging to completely plan an entire economy.
https://chris-said.io/2016/05/11/optimizing-things-in-the-ussr/
Of interest to folks looking at this post -- Chris Said has a deeper dive on US air quality metrics and their provenance, https://chris-said.io/2023/06/19/the-air-quality-index-doesnt-make-any-sense/.
The attempt to make things easy to interpret using a nice relative scale ends up in the end make them very opaque to interpret the number itself in any meaningful way.
>Turned out not to work, probably b/c you can't stablize an inherently unstable complex system.
Or because they didn't have enough computing power at the time
https://chris-said.io/2016/05/11/optimizing-things-in-the-ussr/#:~:text=Leonid%20Kantorovich%2C%20inventor%20of%20linear,1975%20Nobel%20Prize%20in%20Economics.