It doesn’t need nearly as many. AI inference is orders of magnitude more expensive than a single search query (ignoring the fact that Google does it’s own inference with search queries now). And that doesn’t even include training, which is stupidly expensive to do.
It does not use AI data centers. Those are not remotely on the same scale as what we called a data center before this LLM hype started. This infrastructure will not have another use once the bubble pops.
There is a difference between using X amount of resource to provide an actual output that justify spending X amount of resource, and using X*1000 resources to provide zilch.
Citation on it being zilch because they do a fair bit of actually useful work these days while the search engines mostly seem to be going downhill (Kagi being an exception)
Search engine going downhill is mostly a side effect of AI enshitification getting in the way. Even google, which people keep saying is “dying”, works as well as before as long as you’re not using their default page which is filled with AI garbage and other automated content.
Also, Large AI models from large provider doing a fair bit of actually useful work would have to be confronted with other way to do said work. People burning tokens to edit PDF isn’t exactly that efficient compared to opening a god dawn editor.
Are you under the impression that a web search does not use data centers?
Google in particular has a lot of infra to support their search, which actually used to be good
It doesn’t need nearly as many. AI inference is orders of magnitude more expensive than a single search query (ignoring the fact that Google does it’s own inference with search queries now). And that doesn’t even include training, which is stupidly expensive to do.
Sure. But let’s not pretend Web search is innocent here. Wanna be eco friendly, walk to your local library.
Using this a lot lately, huh
It does not use AI data centers. Those are not remotely on the same scale as what we called a data center before this LLM hype started. This infrastructure will not have another use once the bubble pops.
It was said that one was built in 2022. That’s long before they started doing gigawatt scale bullshit.
There is a difference between using X amount of resource to provide an actual output that justify spending X amount of resource, and using X*1000 resources to provide zilch.
Citation on it being zilch because they do a fair bit of actually useful work these days while the search engines mostly seem to be going downhill (Kagi being an exception)
Search engine going downhill is mostly a side effect of AI enshitification getting in the way. Even google, which people keep saying is “dying”, works as well as before as long as you’re not using their default page which is filled with AI garbage and other automated content.
Also, Large AI models from large provider doing a fair bit of actually useful work would have to be confronted with other way to do said work. People burning tokens to edit PDF isn’t exactly that efficient compared to opening a god dawn editor.
derail
Government propaganda making people think it’s AI chat bots why all these data centers are being built
They are to process the enormous amounts of data the government, flock and the rest are gathering to monitor citizens