Google started slipping when it began assuming that it knows better than you do. For instance, when the exact search operator (" ") stopped working.
I'd take the Google of 2014 over the Google of 2023 without thinking twice, and I think that I'm not alone in this.
[verbatim mode] https://www.google.com/search?q=unexpected+newline+in+compos...
[quotes] https://www.google.com/search?q=%22unexpected+newline+in+com...
The top two results are the same, but besides that verbatim mode seems way better to me.
(said the independent search engine creator with the same air of righteousness as when Plato argued philosophers should be kings)
Same situation with python honestly. No I don’t want to use pandas, if I wanted better data wrangling I’d just wrangle in base R then import the wrangled data into python if need be instead of adding another dependency to my python code. Apparently this makes me a heretic based on google search results.
Since I know the top results are mostly useless spams on some searches, I used to go to page 4 or 5 to find the actual results. Now to achieve the same result I have to click and scroll 5 times.
It still works.
Works exactly how quotes used to.
Actually, it has happened with results that include hacker news discussions. I scrape through and expand EVERY comment on a 6 month old post and still find no partial or exact matches for my search keyword. It is very clear to me that Google has switched behavior. It used to flex on how awesome it was to bring you to an esoteric corner of the internet matching your exact query. Now, it tries to converge your queries into what it thinks you want(not what you asked for) and funnel you into large websites at the expense of utility.
Which can be incredibly frustrating if you know the problem well enough to need a different answer.
I also love when my search is narrow enough that Google instead starts to return more generic pages for totally different programming languages instead of the smaller result set that matches exactly. It is very easy to observe as you explore a topic. At some point, you got a wall where any increase in search specificity seems to yield worse results than queries that helped you get to that point.
Which... Why isn't Google learning that some users are constantly seeking out verbatim for results they'll click on and adjusting their search results to be more verbatim?
Some people admittedly expect pretty weird websites to exist. Like who would even publish a good objective comparison between products in a specific category. But people make those queries looking for those types of results, and spammers generate such content to make bucks off the traffic. 20 years ago you'd have gotten nothing in most cases.
Consider coffee. You'll find hundreds of articles comparing the most high-end grinders. You even have PhDs writing white papers on the physics of coffee and how particle size distribution achieved by different grinders affects extraction quality. Those are genuine opinions from aficionados.
The same is true for all kinds of other things. The internet has a lot more than the auto-generated marketing crap you give it credit for.
Sniper in mahwah for example will discuss some of this for example from an HFT angle.
Its results are superior, with way less spam and SEO optimized content.
a) there is no giant industry optimizing for ranking on Kagi. There are whole armies of people trying to game google search, meaning whatever heuristics they have would probably be rendered useless shortly if google were to adopt them.
And
b) Kagi has a much smaller, more technical userbase. So like the early web they can get away with a search experience tailored for those kinds of people and not millions of people powering on a smartphone as their first computer.
Kagi works for you, with your input. It doesn’t do “advanced guessing” by implicitly and opaquely processing your interaction with it.
I guess their "Universal Summarizer" is a part of their indexer, so they try to understand what you want and provide you more relevant results.
They probably have other signals and systems to detect spam and other unwanted things.
Combine it with personal website ranking and lenses, it surpasses Google easily.
Also since it has no ads, there’s no noise.
I don’t use Microsoft products, so I can’t comment on Bing.
I suspect this is a niche feature for power-users only. HN is filled with power users, so they love Kagi, but it doesn't necessarily translate to the masses.
I think it’s important to measure and quantify these things.
I can't provide you with side-by-side examples because I don't compile lists of such examples. I'm searching to find stuff, not researching search quality. I just know that I have more problems getting good search results from Google over a couple of the other engines.
If a user of service thinks a different service is better, then it is better for them by definition. "Better" is a subjective thing. If X seems better for me than Y, then it is.
"Are US drivers licenses forgery-proof?" and "Can US Real-ID drivers licenses be forged?"
gave utterly irrelevant #1 or #2 hits: nothing to do with US or Real-ID (zero mentions), it's about New South Wales, Australia (lots of mentions of "NSW"). (although from US magazine Wired.com, but Google can still tell this article is totally irrelevant).
2) I previously in 2022 posted "Google search relevance fail: result for “Africa longitude” [https://news.ycombinator.com/item?id=30337563] where the #1 site hit and text result were the grossly misleading "Africa/Coordinates 8.7832° S, 34.5085° E", which is not a range, or even a centroid, just a junk location in southern Tanzania, where a (Russian?) SEO spam person seemed to have dropped a bogus pin, mislabeled "оф. 3, 27 ул. Хабаровская, Bila Tserkva, Russia". In Tanzania. 17 months later and same gloriously unfixed garbage results.
Other bizarre top hits were the small Ottawa e-commerce digital-map site https://www.mapsofworld.com/lat_long/africa.html, with the incomplete and inaccurate non-sentence "Africa's latitude and longitude lies between 9.1021° N, 18.2812° E." (That e-ecommerce store seems to have an autogenerated page with incorrect information on latitude and missing longitude for each country).
These are cases where Google itself 5 years ago would have done better. Or compare to Kagi.
So what do you think, will be the goal of all advertisers?
To get their recommendations baked into that AI. Direct bribing/revenue sharing with the manufactur of that AI. Or sneaking data into training sets via bribed employes and so on.
I don't trust black boxes and never will.
Paid search is the future. Either you pay a subscription fee to something like Kagi, or you pay for hardware and electricity turn run your own "AI agent" locally.
It's a negative-sum arms race, but if Google (or whoever) wasn't spending the money to fuel the arms race by passing on ad money, who would pay for Google to exist in the first place?
That's just my hypothesis, of course. I don't know for certain. But it does make me very skeptical about using AI to help search the web.
I want a robust set of search modifiers instead.
How long do you think it will be before that software agent's output will be just as riddled with commercialized content? I think we might have only a short window where we get to enjoy the fairly unadulterated output of the base LLMs by the large providers (yes, I'm aware that they're screening for safety already, but there could be a lot more coming). Imagine a dialogue like this: Q: What should I watch out for when buying new car tires?
A current answer might be something like: Here's a list of things to watch out for when buying car tires: A, B, C, D, E. Remember to always consult with specialists. Shall I explain more about the topic?
A future answer might look something like: According to experts like the professionals at [tire shop near you], here are the things you need to watch out for when buying car tires: A, B, C. You can use this coupon to get a 10% discount off your next purchase at [tire shop near you]. Shall I book an appointment for a free consultation at [tire shop near you] for you?
I'd love to have one, but do any exist? SGE works on google's behalf, collecting data for google who will use what they learn from it against me. I'm guessing that until we can use common hardware to train/host an AI agent ourselves, we're stuck with terrible search results unless we're willing to hand over more and more data to others who want to exploit us, while pretending we can trust them to return results that are in our best interest instead of theirs. That's not a great future to look forward to.
I've had "Search Generative Experience" (AI/LLM synthesized result/summary above the standard web results) turned on for awhile. It is definitely the future. To survive the flood of revenue optimized content (recipes being a familiar example), you need an active [software] agent working on your behalf to read content and make one straight thing out of the crooked timber of the web.