Big Bang: The Origin of the Universe by Simon Singh.

Flies easily fly through the window if it's substantially brighter than the room.

Not in my experience. The walls were dark. And the window was bright occupying half of the surface of one wall. They still flew in circles.

I think it might be what other commenters are pointing out. That flies consider the brightest direction to be the sky, but if it's to the side (with respect to gravity) it messes up their navigation.

I wonder now, do flies ever venture inside of caves?

The sponge design of SHA-3 makes it a CSPRNG basically. You can keep draining it after you fed the data in.

You can do something similar with any cryptographic hash; just feed a (say) 128 bit counter through the hash function, and collect the outputs.

That's a great poin, even better.

It's already happening. For example, ECMWF provides experimental forecasts by their own deep learning model, plus models from Deepmind, NVIDIA, and Huawai.

They perform several model runs with slightly different configurations and then compare the predictions to a reference period, currently 1991-2020. When they say "above average", most model runs fell in the upper tercile of the reference period and the probability is derived from the fraction.

Thanks, is there a summary of this methodology somewhere? Some established climate forecasting textbook? A good literature review?

If anyone is interested in these seasonal forecasts, all major meteorological institutes upload theirs to https://www.wmolc.org, with new predictions coming out every month.

You can get technical ENSO specific forecasts here:

https://www.cpc.ncep.noaa.gov/products/analysis_monitoring/l...

Which you can find from here which has a bunch of other information:

https://www.cpc.ncep.noaa.gov/products/precip/CWlink/MJO/ens...

Which you can find from the forecast tab from the main landing page for ENSO on the NOAA PSL site:

https://psl.noaa.gov/enso/

Up in the Old Hotel by Joseph Mitchell. I finished it yesterday.

It's a collection of articles the author wrote for The New Yorker around the middle of the last century. The stories are portraits of weird, eccentric people and peculiar places in and around New York. Mitchell describes his subjects with dignity, without putting them on a pedestal.

The author is a very good writer and his stories are enjoyable, but only after I had read the book to the last page, I understood what makes this book so great. The stories tell you as much about the author as they tell you about the protagonists. I started the book to get a glimpse of the old New York and finished the book with a great curiosity about the author himself.

If I understand it correctly, you are only attending preceding tokens in your paper. Can the constant bias matrix be made symmetric for unmasked tasks?

They still publish all their data free of charge and are required to do so by law. For weather forecast, they dump all parameters of their recent model runs.

So I guess I'm saying that you can already build such a system yourself quite trivially...

Like Meteocool? https://github.com/v4lli/meteocool

Developed by a former colleague of mine. Open source and my favorite for 2h forecasts about upcoming rain in Germany.

The app „Tiny Weather Forecast Germany“ (https://codeberg.org/Starfish/TinyWeatherForecastGermany) seems to exactly make use of that. There are probably others, but it is the only one I found on F-Droid.

The expensive stuff still bears the trademark of the brand, the lead designer, or the collection. It's more subtle, but ultimately the reason why people buy it. And some brands really just sell the logo at a high price, like Hermes or Louis Vuitton.

Expensive Hermes stuff doesn’t tend to have logos. In fact, not many of the cheap things have logos either.