The existence of offline models highlights a nuance that some people deny even exists though, causing people to talk around one another. I wish it would be more widely acknowledged, as it would make some around AI conversations easier.
Comment on Tankie
starelfsc2@sh.itjust.works 1 day ago90% of people do not use offline models, especially for everyone doing ai code and video. The offline models are undeniably worse and slower. These ai companies didn’t just magic billions out of thin air, most people are using the massive data farms. Also people are generally not playing 14 hours a day maxed out gaming, where for ai they might use it all day during work.
ClamDrinker@lemmy.world 1 day ago
bbboi@feddit.uk 1 day ago
Does the physical location of the hardware really matter?
ClamDrinker@lemmy.world 23 hours ago
It does not but that wasn’t my point. It was that that not all forms of AI usage are the same. The same way someone driving around an EV that they charge with solar power isn’t the same as someone driving a 1969 oil guzzler (or something equivalent). Local usage more often than not means efficient models, low energy consumption, and little difference to other computer tasks like gaming or video editing. But when the conversation is around AI, there is always the explicit expectation that it’s the worst of the worst all the time.
starelfsc2@sh.itjust.works 1 day ago
So I did some more research, and evidently if you’re going to use AI at all, you’re probably increasing your energy usage by using it offline if you use it often (unless you are using renewables), since the data centers generally have cards specifically designed for AI. I think it might just be a case of everyone needs to use it significantly less, it’s like if 4k gaming was something the average joe was doing. If everyone was doing that 10 hours a day, we would have a big problem.
It’s kinda like saying it’s not immoral to go for a pleasure drive, but if you’re driving around 10 hours a day that’s probably not good and you should minimize it as much as you can.
ClamDrinker@lemmy.world 23 hours ago
That’s pretty interesting. And I totally agree with your last part. One counterpoint I would have is that local models are often more efficient though, and there’s very little checking you can do on how much your query actually costs in the cloud, while using it at home you can monitor your GPU usage and your power bill. But yeah at the end of the day using it as little as possible is a good habit.
brucethemoose@lemmy.world 5 hours ago
I am late to this argument, but data center imagegen is typically batched so that many images are made in parallel. And (from the providers that aren’t idiots), the models likely use more sparsity or “tricks” to reduce compute.
Task energy per image is waaay less than a desktop GPU. We probably burnt more energy in this thread than in an image, or a few.
And this is getting exponentially better with time, in spite of what morons like Sam Altman preach.
There’s about a billion reasons image slop is awful, but the “energy use” one is way overblown.