Loved the writeup!

Found the manual latent space exploration part really interesting.

Too many LLM/diffusion explanations fall in the proverbial “how to draw an owl” meme without giving a taste as to what’s going on.

Turning off the scroll mode worked very well for me on a mobile.

Even if one paid monthly, why would they actually stop the data collection?

Well YouTube offers a no ads version for money. I personally don’t see a realistic alternative to ad supported social media so you’d have to ask someone who does think that.

Probably should also explain how to use a book…

You’re talking about ancient technology here…

IRL I’ve seen similar discussions devolve into an hour long bike shedding meeting about how to define thresholds for warnings, track new ones, etc.

Before the end, I had them all fixed. Zero is far easier to deal with…

Once got one of those with a disclaimer that clicking any link was giving permission to subscribe me…

I believe they included the “unsubscribe” link too…

How did OP debug this without a serial port?

I’ve messed around with porting coreboot on two desktop platforms but always had the benefit of a HW serial port…

because the x280 and x270 are similar enough I didn't need to try very hard to get it to post or boot a live USB to further investigate (effectively acting as a decent template for me to work off of)

The console viewing itself was provided by `cbmem -1`, which I could run via a NixOS live USB with nixpkgs#coreboot-utils

You can sometimes find the serial lines if you are careful. Otherwise you can use the flashrom to store the output, and read it back out after each failure. It is much easier to just poke around and find the serial if you can, either from schematics (it seems the author has these) or by hand with a lot of patience or board scrying.

Sorry if this sounds harsh but this looks similar to commercial weather services in terms of the types of layers. How does this actually work?

Did you manage to track down their other services’ providers and negotiate access?

They can just use the free government services. Most of the 3rd party services just wrap the various government services behind an API.

This is true for many free weather APIs out there. But our weather API is global so we’re not just passing through a single government feed. There’s quite a bit of engineering to normalize all of the data sources. Japanese alerts for example are just XML feeds. All the sources update on different schedules, and have different forecast ranges, so it's a challenge to make sure the best forecast is returned at request time.

Some of our data is proprietary. The lightning strike data is from our own lighting detection network. We also offer the only hail forecast model available which was developed in house :)

It's a fair question. There's a lot of free sources of data out there. This app is powered by our own Weather API: https://www.xweather.com/weather-api

The data is aggregated from global meteorological offices and other sources (some free, some paid). Some of the datasets are proprietary to us. For example, the lightning layers come from our own lightning detection network, and certain forecasts (like hail) are developed in-house.

This app was built to showcase our data. The vector mapping tools are also available via API. https://www.xweather.com/mapsgl

I also wonder about the state of packet radio.

When I had time, I had no money for equipment. Now that I have money and knowledge, no time…

I don’t think the current cost structure of software development would support a professional engineer signing their name on releases or the required skill level of the others to enable such …

We’d actually have to respect software development as an important task and not a cost to be minimized and outsourced.