This is a screenshot of my weather app showing forecasted (smooth blue/purple) rain of the French Arome model VS rain radar (sharp green/yellow contours). It actually does a pretty good job on the large scale but the details aren't always quite right. Dies anyone know how rain probabilities are computed from these models instead of fixed predictions?
Agree. The issue with reproducibility comes up mostly once you are building your own containers, many of them, and run them in production but then it can hit very hard because it becomes too expensive then to work around all of the following: Once you are hit by a problem (sometimes years after building the original one), it's hard to debug bc you can't be certain to be able to reproduce what you have. It's even difficult to just find out what is in an existing Docker container, companies are running costly scans for that. And finally you can't compose (you can mostly layer) as well one Docker container with another. Using Nix with flakes gets rid of all of that and you can still have Docker or other containers as output.
Cryptpad is using a custom, zero-knowledge Version of only office. They describe the relationship between both here: https://docs.cryptpad.org/en/FAQ.html#what-is-the-relationship-between-cryptpad-and-onlyoffice
But can you really get an autoimmune disease by eating too much supplements? I thought the body would just not retain them. It'd be quite difficult to eat exactly the quantity required without any self regulating mechanisms in place.