Ame on Nostr: I've talked to several very skilled CS/AI people in my lifetime, and they all agree ...
I've talked to several very skilled CS/AI people in my lifetime, and they all agree that search engines died around 2018. I don't know if they all knew why, but we know why - there was a huge push to get rid of the old algorithms that optimized, quite spectacularly, towards giving users what they wanted.
People have talked about using LLMs instead. They can certainly do a few of the things that Old Google could, but there are some things that they can't manage, and probably won't ever manage. They can walk you through the trivial stuff, but falter whenever they face something that the meta-learning inherent to their training process isn't capable of grasping with limited data.
Some people have tried to make their own, like Brave Search. I get the feeling that these are half-assed efforts, without a real team or any real resources behind them, but even so, I think you had to have scale to get the old algorithms to work. Might be like peak oil - no way to get back once the initial route has been run.
Published at
2023-06-15 09:30:40Event JSON
{
"id": "6625f97b9bc11a4f49d7e107e8e8eeacccf54fd13618c992154de78e60e4de03",
"pubkey": "7271f65788cf3d629727a54de8a74cfaa69f3ee1232c7dfc6bcb8059d28467e1",
"created_at": 1686821440,
"kind": 1,
"tags": [
[
"p",
"748fa8a28d3a079b7650de9088739a8ce36fdbd6d83b004565b1609787bde390",
"wss://relay.mostr.pub"
],
[
"e",
"940d48580d32a48f50a7849ff7288df240f658328b381b6a3c843522f79b5752",
"wss://relay.mostr.pub",
"reply"
],
[
"mostr",
"https://poa.st/objects/9c523a60-78cc-468a-b0d4-157fb8405215"
]
],
"content": "I've talked to several very skilled CS/AI people in my lifetime, and they all agree that search engines died around 2018. I don't know if they all knew why, but we know why - there was a huge push to get rid of the old algorithms that optimized, quite spectacularly, towards giving users what they wanted. \n\nPeople have talked about using LLMs instead. They can certainly do a few of the things that Old Google could, but there are some things that they can't manage, and probably won't ever manage. They can walk you through the trivial stuff, but falter whenever they face something that the meta-learning inherent to their training process isn't capable of grasping with limited data.\n\nSome people have tried to make their own, like Brave Search. I get the feeling that these are half-assed efforts, without a real team or any real resources behind them, but even so, I think you had to have scale to get the old algorithms to work. Might be like peak oil - no way to get back once the initial route has been run.\n\nhttps://i.poastcdn.org/f33165d3f61671fb9b6985d4bbb971d4be90060f5a15adc26beecced6da3251b.png",
"sig": "b627569dac9909cf349a3d3cb8cb4615fbbc0bf61a2cc551822808f4a9a8b6b67565f0f86b155d77564b8142948edffce7c6affb542235252ef7f4804bf2d701"
}