That is a very fair point! I just ran a not very scientific benchmark with the system under load, and posted the raw logs in a sibling comment above, but the short answer is that it's hitting 11.94 tokens per second for generation - while it's also being a binary cache and CI build server.
Totally just vibes based, I think it goes up to 20+ tps when it's not under load (and that's me trying to be conservative). For context, reading speed at 250 wpm would be around 5 to 6 tokens per second.
I wonder if there actually does exist updated to-the-minute imagery of various places, just not from sources publicly available on platforms like Google Maps?
Yes, you can buy it from commercial providers like Planet or even Airbus. They update a few times a day, though depending on the place you are interested in, you may need to put in a request for them to image it.
Its actually not that difficult. I used to fly a satellite that could photograph anywhere in the world at least once every few days.
It must be quite valuable data. One anecdote I heard/read somewhere was that firms often use satellite imagery of parking lots (over time, as one measure among others) to gauge how popular a place is. I don't know if it's true or not.
They have the resources to spend (to them) a small amount to confirm/check a play before they make it, it's been a thing since commercial satellite imaging became a thing.
Hell the (somewhat terrible but somehow enjoyable) TV show Billions had it as a plot point years ago.
Similarly, I listened to a talk from someone who used to perform analysis of aerial images of farmland to estimate yields at harvest, which would then be used to trade in the futures market.
Quite an interesting read. I wonder how well this system would work if you used something other than Midscene.js to fingerprint the models? It was limited in scope on how the models interact with the sites for study's sake, but would be interesting to look at further.