Showers and thunderstorms!
Comment on Hope it doesn't rain
empireOfLove2@lemmy.dbzer0.com 3 weeks ago
Who the fuck put an LLM in my goddamn weather app?
basxto@discuss.tchncs.de 3 weeks ago
Comment on Hope it doesn't rain
empireOfLove2@lemmy.dbzer0.com 3 weeks ago
Who the fuck put an LLM in my goddamn weather app?
Showers and thunderstorms!
boonhet@sopuli.xyz 3 weeks ago
Idk, this is the kind of thing an LLM would actually be good at. I think this is much simpler automation at work.
lmuel@sopuli.xyz 3 weeks ago
But it’s like… a couple of numbers vs more text. I reckon looking at the numbers is a lot quicker lol
boonhet@sopuli.xyz 3 weeks ago
Oh that’s why I get my weather from yr.no. Nice table, hour by hour for next few days and ranges for the later ones. I also mostly visit from the desktop website so this is what it looks like to me:
Image
(I’ve even learned a few words in Norwegian from it since I’ve never bothered to switch it to English)
The one in the screenshot, I’m assuming, loops through preset ranges with 4-5 PM being a special one because that’s around where people start going home from work, and just gives you a string based on the weather conditions. So morning till 4, 4-5, then 5 and onwards.
Now if you actually wanted to do it with an LLM (not that you should, it’s pretty wasteful if you consider that it’s pretty much a prompt per every person looking at the weather, maybe you can cache for an hour per location, but there are millions of locations), you’d give it the dataset for the entire day and, being an LLM, it’ll at least make the text flow much better than it is on the screenshot. An on-device tiny LLM as an opt-in feature would perhaps make sense here, but as you go smaller, the quality goes down too, so again not much point IMO.
Prok@lemmy.world 3 weeks ago
Not sure why you’re being down voted… Pretty sure you’re right…
Scribbd@feddit.nl 3 weeks ago
I can say that it is as you say. I don’t know this app, but ours takes the values of the forecasts which are defined in timespans. The backend loops over that and matches it to texts to construct a blurb like that.
It could simply be that the meteorologist defined a seperate period as more intense or is associated with a weather warning. And the simple automation of this app doesn’t check for continuity.
To be fair, I am unsure if ours does this as well. I don’t work directly on the app. I am an infrastructure guy. I do know it is checked by a human meteorologist before updating or publishing.