Hacker News Reader

Comment by tines | Hacker News Reader

I just performed that search in Google and the top two results are links to PyPI and Snyk packages for micropython-html.parser. The third result is a link to python.org documentation for html.parser (not micropython).

Bing returns the result for PyPI and the one for the non-micropython python.org result.

I don't see a big difference here (except maybe me learning about Snyk through google and not Bing?).

View on HN

I mean, this is a big part of the issue. Of course the issue isn't reproducible; Google is trying to do the search based on their model of what you want back. Sometimes you benefit from that, other times it pushes the results you want way down. I just tested this and there were some micropython related results in the top ten, but the majority were for beautiful soup with no mention of micropython.

> Of course the issue isn't reproducible; Google is trying to do the search based on their model of what you want back...

There's a problem here though, which is that there is no way for me or any other reader to discern whether the reality of the situation is "it's not reproducible and the commenter is misremembering basic facts about their search results due perhaps to some unrelated frustrations" and "it's not reproducible but the commenter's account is accurate."

I completely share the views about Google's search results being terrible, but it's still possible to overstate how bad they are. And I do find it fairly difficult to believe that Google would show some people paid programming courses as the top 4 results for that query when much more relevant results clearly exist and everyone attempting to reproduce this gets those more relevant results.

nor are we anywhere near being able to inspect how google is doing all this.

if only there were some way to somehow be able to share that information with everybody, some technology so that we can all access such potentially useful information about how google is functioning right now

sorry about the snark, but it's a vent for accumulative frustration from seeing a worsening trend in this regard.

Kafka would recognize this trend; computing is truly becoming a "bureaucrat's best friend".

I believe the OP because I've had that exact problem once myself, except in my case it was "helpfully" substituting FreeBSD in the query with Linux.

I often find that frustrating about Google search, but I'm also unconvinced it's a problem that a more-AI-driven approach will avoid. Seems like the idea to try to convince the model to interpret things differently based on user history is still likely to get pushed by product managers, and still technically capable. And then "no tell me about the framework" "conversationally" vs tailoring Google search terms seems a bit of a wash.

I'm about 80% sure that the reason why Google became so bad for my searches is that they added AI elements to it, to make it try to figure out what I'm searching for instead of taking my word for it.

In that sense, I'm not sure more AI would improve the situation. OTOH, better AI just might. There doesn't seem to be a "no AI" option on offer.

Based on my usage of ChatGPT, I think that a more AI driven approach could help a lot. With ChatGPT if it misinterprets what I'm asking for, I can make a clarifying followup query.

> Of course the issue isn't reproducible; Google is trying to do the search based on their model of what you want back.

There could also be other things that make it non-reproducible as well.

It sometimes feels like search just returns whatever it has on hand that seems somewhat similar to your search. It may then do an asynchronous request to pull into cache results from a deeper backend search, but by the time the frontend gets them you've already been served your results. Subsequent searches from other people for the same terms would then get better results than you did, so this sort of optimization would on average improve search quality even if quality is poor for the first searcher.

I have no idea if search engines actually do this. But it seems to explain some of their more mysterious behavior, like immediately returning a bunch of results that have nothing to do with your search terms.

That sounds plausible. There could be some service aggregating search results from multiple systems, and if one of them is slow (maybe breaking some SLA), it would be omitted.

I guess there are probably a hundred Google employees reading this who know whether it’s accurate and we’re just speculating.

> I just performed that search in Google

It's been publicly known that Google has been bubbling search results for years now, and that one individual's search results have no relation to another's.

Every individuals everything has no relation to another's - which I think is a core cause of society's disconnects. Not that people got along before algorithms drove everything, but it's certainly been a destabilizing factor.

This is a good argument for LLM search: have search be conversational, and let you say "hey, I actually meant this!". Then you could have different chat environments, each tailored to micropython results, CPython results, IronPython results, etc.

Of course, how can you scale this for every user?

And the responses will be:

"No, you didn't" "I didn't say that" "Seems like your are using on old version"

I think it's more that Google has gotten noticeably worse, not that Bing has gotten better.

All you need is for Google to get worse and Bing's marketing to get better (which it has, by nature of being associated with the biggest tech hype train of recent memory).

I also get the same result when logged-in and incognito. Others in the thread seem to get them also.

Google can incorporate quality signals in near real time. Those signals include leaving and immediately returning to the SERP. So us discussing these particular results could have already influenced them.

This comment thread is now #6 and #7 for me

Freshdocs is incredible. A lot of people are stuck in the old mental model where Google released a new index every month.

That's actually super impressive. So there's basically no "index" anymore, in the sense that the results returned are almost "dynamically" generated? And I'd guess that all of that still needs to be pretty cheap, computationally speaking.

I've seen this before - I've complained on HN that google has no results for X, and within two hours google's first(and only result) was the very thread of me complaining about it.

Google knows it's you if you did all of that from the same computer.

Google is more than capable of doing such basic fingerprinting, but do you have evidence Google uses it for search results? You can do a trivial experiment on another Google platform: open some video on YouTube, leave it, and look at the front page. Then, look at the YouTube front page in a private/incognito tab. (Alternatively, use two different private sessions.) In my experience, no matter what device or network I use, and no matter what the video was, the first page will always show videos related to the one I opened, and the second will always show an extremely generic set of clickbaity videos (likely based on GeoIP). This suggests that Google uses ordinary cookies for basic relevancy ranking.

Considering people switch to Incognito Mode to get different results, it'd be weird and not particularly commercially savvy to frustrate them and feed their paranoia by serving them the same ones. Non-Google search engines and browsers exist.

I'm also reminded of a CTO complaining that his test sessions suggested the default Google ads on a page with little text content all related to dating. We pointed out that that from Google's point of view, this was probably a sensible ad-targeting decision for a user running an Incognito mode browser (or any browser without tracking cookies)...

> Considering people switch to Incognito Mode to get different results, it'd be weird and not particularly commercially savvy to frustrate them and feed their paranoia by serving them the same ones. Non-Google search engines and browsers exist.

I just installed Google Chrome on a fresh Windows 10 VM to repeat the experiment outside of Incognito Mode. First, I opened a video. Then, I checked the front page and confirmed that related videos had been added. Then, I closed Chrome, reset the VM's state to before I had visited YouTube, and checked the front page again. (I don't think this scenario would be too indicative of a prying user; I can easily imagine corporate systems that regularly wipe out browser data.) Again, the related videos were replaced with generic videos.

I then tried the same thing, except by clearing browsing data through Chrome's UI instead of resetting the VM. The results were the same.

Obviously, these observations could all have been manipulated by sufficiently conspiratorial fingerprinting. But the simplest explanation, in the absence of good evidence to the contrary, is that the site uses and respects browser cookies for its recommendations. Thus my request.

Let me guess you are using chrome?

I have no idea why Snyk isn't treated as a spam domain.

Reply on news.ycombinator.com