# Google scums

## Concept

You have perhaps found internet boring lately?
"Google can't find anything".
A LLM told you "you are not imagining things".

### Sick days.

Well, this page is my take on it.

## Full disclosure

I did 6 interviews at Google a long time ago. The tech part was OK.
But one guy did not like me. What not to like?

I don't hate them, though.
But I do believe they are corporate scums.
Even my friend working there ;)

## A normal search

You do a normal search and you get...
The same shit again!

So it will be a mix.

- the AI
- a bit of youtube
- reddit
- wikipedia
- amazon
- if you are lucky one or two good pages

And that sucks because wikipedia is nice but complementing more opinionated websites.
You need anchors, you need passion, both.
The AI is wikipedia summary, it's the main attraction and it fails to deliver.
Reddit is one guy, 2 answers, or 200 if they angered people by being themselves.

But let's hack it. Test, probe, see what is really going on.

## A good search

A good search to see if the search engine is apt but diluted by the unfathomable size of the Internet.
So we'll search that : soviet camera repair

Are you a commie? No, I am anarchist.
Soviet is really taking away most chunks of the web space. It is a place, a time, an industry, military.
It is not like electronics : millions of products, billions of sales from the vintage Marrantz to the latest GPUs.
Search soviet on amazon, it struggles. A few poster of Che Guevara, perhaps. A Робот perhaps. Nice shit.

Then you add camera. Well soviet camera, it's oddly specific.
There is a Moldovian joke a Moldavian teacher told me.

- A guy from Moldavia visits Russia.
- The guy presents the factory : here we build the biggest tanks in the world
- Then showing the port : here we build the biggest warships in the world
- etc.
- Then he shows the electronic factory : here we manufacture the largest chips in the world

Anyway. Stop interrupting me with your jokes.

So it's not Japan camera, soviet camera, I mean they built 10 or so. 1 model to fit them all!

And then you end it all : repair. No sales. We are super clear here.
I don't know google search internals. But in essence, we are super nice with the Vector-space partitioning or equivalent.
The idea is as old as b-tree, binary trees.

You have a large group. You can discriminate half of it. If you do that a few times, it exponentially reduce your search area.
Please note the effort below to illustrate. Hand-made. So you discard B on the right, then A1 that is half of A on the left, etc.

```
┌────────────────┬───────────────────┐
│                │                   │
│                │                   │
│                │                   │
│                │                   │
│       A1       │                   │
│      1/4       │                   │
│                │                   │
├────────────────┤         B         │
│                │        1/2        │
│      A2.1      │                   │
│      1/8       │                   │
├────────┬───────┤                   │
│  1/32  │       │                   │
├────────┤  1/16 │                   │
│        │       │                   │
└────────┴───────┴───────────────────┘
```

And that's if you partition in 2. Because soviet is not half of what Google indexed, right?
Nothing is half of it, but it is specifically rare, and very rare combination.
So camera and soviet already rare, reparation is super niche.

If you'd search Japanese camera, you'd get merchant selling some. It is still to buy, you may want some.
Here, with that search, I cut in the space violently. I am not searching to buy something. I am not searching for any camera.
I am not asking for an explanations on the USSR. Soviet and repair, camera.
So it is a clean test, it really pushes the search toward cutting through the data and getting relevant results.

Let's [search it](https://www.google.com/search?q=soviet+camera+repair)

![Google search result](google/search.png)

You see on the right? Random sponsored Japanese repair shop. I checked, Japanese brands, nothing soviet.
Lower than the screenshot, freaking Amazon.
Facebook, youtube, poor, unrelated, it is trash.

Not even a third of the links are related.
I admit that this one is incredible : [learncamerarepair.com](learncamerarepair.com)
Real soviet era repair manual for cameras. And it is like a small part of their site. The thing is incredible.
Bookmark. When was last time you bookmarked a new website?

So let's continue. Go to page 5.
There is nothing to see.

- [Telefonica Deutschland](https://www.telefonica.de/) : German mobile operator
- [Instagram](https://www.instagram.com/reel/DaND0Luh2n2/) : the guy is testing retro Japanese stuff
- [Australia Space Agency](https://www.rnz.co.nz/news/world/680057/australia-s-space-agency-says-beach-objects-likely-fuel-pressure-vessels-from-launch-vehicle)

It is trash, it is not related, it is only established companies. It is paying or SEO.

## The issue

Well, we can conclude confidently that Google does not give us an answer that feels like the world wide web.
Nothing in Russian. No old abandonned websites. Little passion, exponentially decreasing quality of the results.

We can see that the google :

- Giving priority to unrelated sponsored content
- Mediocre and biased first page
- Trash after, absurd at page 5. Australia Space Agency... why?

So there is 2 problems we can infer :

- Retrieval has issues
- Google favors paying content

But is that everything? We assume that Google indexes everything, it is said to semantically understand pages.
Meaning not just searching keywords, but understanding with a model what the pages are about.

The terms constrain the space:

- Soviet → a geographic/historical/industrial category.
- camera → a specific object class.
- repair → a specific activity and intent.

A good semantic understanding system should recognize something like:
"The user is probably looking for expertise around restoring Soviet-era photographic equipment"

It should not need to decide between:

- a telecom company
- a space agency
- an insta about retro

Those results indicate that either:

- the ranking system failed to connect the query meaning to the available pages
- the relevant pages were not strongly represented in the searchable index

But the query is almost a perfect case for semantic retrieval:

- entity: Soviet Union / Soviet cameras
- domain: photography
- task: repair
- likely audience: hobbyist/collector/technician

The expected results are not hard to imagine:

- repair manuals
- enthusiast archives
- camera forums
- restoration blogs
- museum pages
- parts suppliers

So I doubt it really is a retrieval problem.
I say that it behaves as if the data was not here.
The result entirely collapse. It is not less relevant, it is semantically not relevant!

I did not expect the result to be so bad. I expected in order :

- The sponsored stuff, pay to rank
- The good stuff
- The electic stuff
- The soviet stuff

### It is almost a Fermi Paradox : Where is everything?

I can tell you it's all there.

## Thoughts

If Google's goal is no longer to give you the best results...
If they want you to feed you with easy stuff like google, their AI, their youtube, Amazon...
Why even go crawl and index the web? Why bother?

They may do it, like a tradition, but not as in actively fighting against search engine optimization.
They would do it, at least for R&D, to feed their AI.

But their algorithm and data may be very simple:

- search in a small list of 1000+ websites who give good money
- index a few millions websites, find a dozen of matching wesbite, who cares, and mix them

I don't see in the result, mediocre then collapsing, the result of the indexation of the www.
I can't distinguish the result from what would produce the simplistic thought experiment model I just described with 1000+ websites.

A result that is obviously outside the semantic neighborhood is not merely a disappointment in ranking quality. It is evidence that something upstream in retrieval or candidate selection is not behaving correctly.
This engine did not give what I wanted but what it wanted to give, wether it is money from sites, cheaping indexing, discarding large area of the Web, who knows? But somewhere there, at Google, they shaped what you could retrieve. It looks more like a customized portal than an entry door to Internet.

### They hacked you.

## Solutions

We may go back to meta-crawlers, custom engines. We'll see.
You are already here, neocities. That's a shift. To me too.

### And you may start using [DuckDuckGo](https://duckduckgo.com) and [Wibi](https://wiby.org)

DuckDuckGo? [It worked](https://duckduckgo.com/?t=ffab&q=soviet+camera+repair&ia=web).
I think it worth a shot. It is a different window onto the Internet, a different chunk of it. I like the spirit. But it will not always be the case. Some searches fail on both. Or the the three of them. Plurility is the key. Rather than searching page 2 or 3 of Google, you may want to search page one of three websites.

Now, if you break a leg on another site, I can't really help you.

If you only ever access Internet, the exponentially growing largest network in Human history, through one window that acts as a filter...
Are you still on internet?

### Discounts on red pills.
