5 ms·

That sounds like it makes sense - but if human readers know what's a good search result and what isn't, why would Google not be able to optimize their algo acco

That sounds like it makes sense - but if human readers know what's a good search result and what isn't, why would Google not be able to optimize their algo accordingly? If the results for a query don't match the "nature of the query", it's a bad result. If the algo can be gamed, it isn't aiming for the actual goal, i.e. it fundamentally fails at information retrieval.

Gigachad 4y ago
Because anything google can detect, the spammers can generate to match. We will soon enter an era of GPT generated spam which not even humans can tell if it's real info or just garbage that looks real.
- thr717272 4y ago
  Luckily on Kagi I can just block sites the moment I realize it is nonsense. Or boost or even pin sites that I feel are extra valuable. Back in the old days I had a txt-file on my desktop containing domains with a minus in front, llke -example.com -example.org for all the worst offenders, so that I could post them on the end of every contested search. After a while I think Google found a way to ignore those as well as every other filter I had - at least I gave it up.
  - trinsic2 4y ago
    I have never used anything but google, duckduckgo and startpage. I always end up coming back to google because the results are not as good. Well startpage has the same results as google, but i stopped using it for reasons I forgot. What made you chose Kagi?
    - thr717272 4y ago
      Tested it and found it realized doublequotes something none of the big ones that I can access (bing, ddg, google) does. That was it. And the first time I saw a reported bug in the search results getting serious attention and a fix I was sold.
onion2k 4y ago
if human readers know what's a good search result and what isn't That assumes there's a single, canonical idea of what a "good result" looks like. I imagine that's dependent on a lot of factors, from the context of why the user is searching, their demographic, and their biases. When I search for "chocolate chip cookie recipe" I want straight facts - ingredients and bullet point steps and nothing else. My mother would want some background about the recipe, where it's from, and what makes it different. You can't get that from the search term alone. You need a lot more context. It's exactly what the promise of 'personalized search' based on tracking everyone around the internet told us it would give us but it doesn't seem to work very well.
- Nextgrid 4y ago
  > That assumes there's a single, canonical idea of what a "good result" looks like Maybe we can't agree on what people want, but what I'm sure we can agree on is what people don't want. I don't recall anyone ever saying "I wish there were more ads in the content I'm consuming". So we could at least start with that.
  - bootsmann 4y ago
    Yes, and now turn that sentiment into a metric that can be measured as input to the model. At first it was click-through rate, but that turned SEO into sites into clickbait to get high click scores. Then they switched to time spent on the site, and as a consequence we get recipes that are filled with 2h of prose above them just so people stay longer on the site.
- ajsnigrutin 4y ago
  Recipes are a pain in general, even without google, doubly so for baking. Baking is a science, not art... so recipes calling for a cup of flour and two teaspoons of butter are useless to me, while recipes with ingredients in grams are useless to many american grandmas. On the other hand, a pinterest page with a photo and a large overlay asking me to register is useless to both of us.
  - 4y ago

Gigachad 4y ago

Because anything google can detect, the spammers can generate to match. We will soon enter an era of GPT generated spam which not even humans can tell if it's real info or just garbage that looks real.

thr717272 4y ago
Luckily on Kagi I can just block sites the moment I realize it is nonsense. Or boost or even pin sites that I feel are extra valuable. Back in the old days I had a txt-file on my desktop containing domains with a minus in front, llke -example.com -example.org for all the worst offenders, so that I could post them on the end of every contested search. After a while I think Google found a way to ignore those as well as every other filter I had - at least I gave it up.
- trinsic2 4y ago
  I have never used anything but google, duckduckgo and startpage. I always end up coming back to google because the results are not as good. Well startpage has the same results as google, but i stopped using it for reasons I forgot. What made you chose Kagi?
  - thr717272 4y ago
    Tested it and found it realized doublequotes something none of the big ones that I can access (bing, ddg, google) does. That was it. And the first time I saw a reported bug in the search results getting serious attention and a fix I was sold.

onion2k 4y ago

if human readers know what's a good search result and what isn't That assumes there's a single, canonical idea of what a "good result" looks like. I imagine that's dependent on a lot of factors, from the context of why the user is searching, their demographic, and their biases. When I search for "chocolate chip cookie recipe" I want straight facts - ingredients and bullet point steps and nothing else. My mother would want some background about the recipe, where it's from, and what makes it different. You can't get that from the search term alone. You need a lot more context. It's exactly what the promise of 'personalized search' based on tracking everyone around the internet told us it would give us but it doesn't seem to work very well.

Nextgrid 4y ago
> That assumes there's a single, canonical idea of what a "good result" looks like Maybe we can't agree on what people want, but what I'm sure we can agree on is what people don't want. I don't recall anyone ever saying "I wish there were more ads in the content I'm consuming". So we could at least start with that.
- bootsmann 4y ago
  Yes, and now turn that sentiment into a metric that can be measured as input to the model. At first it was click-through rate, but that turned SEO into sites into clickbait to get high click scores. Then they switched to time spent on the site, and as a consequence we get recipes that are filled with 2h of prose above them just so people stay longer on the site.
ajsnigrutin 4y ago
Recipes are a pain in general, even without google, doubly so for baking. Baking is a science, not art... so recipes calling for a cup of flour and two teaspoons of butter are useless to me, while recipes with ingredients in grams are useless to many american grandmas. On the other hand, a pinterest page with a photo and a large overlay asking me to register is useless to both of us.
- 4y ago