Hacker News Reader

Comment by jmyeet | Hacker News Reader

The "Google is getting worse" trope is popular with a certain anti-mainstream segment but there's almost never examples of how something is wrong or bad or worse. This article has a few examples so I googled them myself:

- "release year modern love roadrunners" -> 1976 (not 1995 as claimed). This article was posted 5 hours ago. Why am I getting the correct results then? For the record, the search "roadrunner modern lovers" (as per the screenshot) also says 1976;

- "brian jonestown massacre anenome" -> lots of links including a video and lyrics (article claims was unknown or no details)

I really don't know why there's so much reality-bending to make this story true.

View on HN

Google results are customized based on location, profile, ad settings, what random experiments you happen to get enrolled in, etc. Reality is indeed being bent, but only inside the personal search bubbles that you, the author, and every Google user lives in.

It's been well known for a long time that Google extensively customizes search results per individual based on dozens of unknown factors. It's entirely possible they get those particular search results right for one person and wrong for someone else, and nobody will ever know why.

Which makes things worse in a way - just like with LLM AIs, you can never be sure that what they give you is actually right unless you already know, so there's much less point in using them at all.

I'm not entirely sure what to say to someone who genuinely hasn't noticed that Google search results have gotten a lot worse in the last 5 years or so, regardless of whether any individual search result was right or wrong.

I feel like we're in year 10+ of this topic coming up to front page of HN with dozens of comments a couple of times a week. I've personally not noticed the degradation but I guess I'm a light google user, mainly just use it get to sites I already know about, I feel like the problem for all these people that are passionate about it is the vast majority of people not on HN are like me.

It really is bad though. I search a lot daily, and every search seems to be 50% pages of this sort: yoursearchterms.info

Because they've re-indexed since then and different hallucinations will exist.

If a given fact is relatively new, I would expect (and forgive) search results to not necessarily reflect that.

But in this case we're talking about ~40 year old song. The reddit thread mentioned in the article is 10 months old.

I've also checked this logged in and incognito and get consistent (and seemingly correct) answers.

So what are we left with? Is the author in some experiment giving bad results? Were there bad results that got corrected between the time of writing and now? Did some Google search quality engineer see this and correct it?

All of these seem like a stretch. I'm perfectly willing to entertain the idea that Google has become worse but there are rarely conrete examples. It all just comes down to vibes. And when there are examples (as per this post), it doesn't match my experience. Can I really be that lucky? That seems to stretch credulity.

Probably the most egregious example of bad search results I've personally seen is how an astroturfed propaganda site was the top result for the Armenian genocide for literally years [1].

[1]: https://www.vice.com/en/article/how-google-searches-are-prom...

I feel like 'vibes' is being used to dismiss anecdotal evidence, which, amalgamated across an increasing swathe of technical users, approaches actual evidence.

The very fact that it was clearly wrong in the example shows you that Google is capable of building a flawed Knowledge graph. This is not vibes, this is either a bug or, frankly far more likely, the inevitable problem of trying to compress all of human existence into a LLM model.

Given that LLM training is an imperfect storage mechanism, is it really hard to believe that a given iteration of the model will just not "know" arbitrary facts.

My personal anecdata on this is that searching for the 'ephemeral port range' the Google AI summary was wrong, even though the Wikipedia reference it used was correct.

Reply on news.ycombinator.com