The 1000 AI Domains that Really Really Really Really Really Want Your Attention (and money probably)

Last week, I saw an interesting entry on the front page of Lobsters. It was a blog post about a math puzzle that the author had found on a telephone pole in the streets of San Francisco. The author had tried to solve the puzzle but eventually gave up and ended up writing a blog post about his experiences. Thinking I would have better luck, I also tried my hand at solving the puzzle, gave up, and am now writing a blog post about MY experiences.

However, unlike the original post, this isn’t going to be a story about math or puzzles but rather something else entirely. That’s because after a sound defeat at the hands of a telephone pole math puzzle, my lizard brain did a backflip and immediately thought it would be a fun idea to try to brute force the solution and see what would happen if I just tried visiting a bunch of numerical .ai pages. So thus began my journey to see every website from 1.ai to 1000.ai.

Background

Before we begin, let’s understand a bit more about the .ai domain. Officially, .ai is the TLD of the country of Anguilla. In practice however, the TLD is more commonly used to create vanity domains showing a site’s relationship to artificial intelligence instead, usually to market a product or a startup.

One thing to note is that .ai domains are generally quite pricey. On anecdote alone, the average .ai domain goes for about $100 to acquire and $100 / year to renew.

Special domains, like common words or numbers, however, can go as high as $10,000 to acquire with the same $100 annual renewal fee.

All this is to say that the sites I was visiting were less like the digital storefronts or houses that people commonly parallel websites with and more like Internet mansions, highly valued pieces of property that some person bought with a lot of money at some point. This made it all the more interesting since I wanted to know what kind of people had bought these domains and what they were setting up behind them.

Planning a trip

My next step was figuring out how to visit all 1000 sites. Doing it manually was going to take forever, so I wrote a simple Python script to iterate through a numerical range. For each number, I had Selenium visit the site and screenshot the page.

from selenium import webdriver
from selenium.common.exceptions import WebDriverException
import urllib3
import time

driver = webdriver.Chrome()
driver.set_page_load_timeout(10)

for i in range(1, 1001):
  try:
    url = f"http://{i}.ai/givemeprize"
    print(f"Trying {url}...")
    driver.get(url)
    time.sleep(1)
    driver.save_screenshot(f'./screenshots/{i}.png')
  except WebDriverException as e:
    print(e)
  except urllib3.exceptions.ReadTimeoutError as e:
    print(e)

driver.close()

One issue with this process was that due to the unpredictable nature of each page I tried to visit, earlier versions of the script would often crash, either due to domains not resolving or servers stalling and eventually timing out the client. I did eventually fix these by tweaking my Selenium configurations and adding error handling, but it was still annoying to deal with these edge cases.

Diamonds in the rough

Eventually, the script did finally visit all 1000 pages. Exactly what did it find? Well, unsurprisingly, most of the hits were either parking pages or generic 404s. However, a few of the pages did catch my attention:

0-9

Nothing very interesting to see here, but it looks like these domains are all owned by a single person. ICANN lookup indicates that the owner appears to be an associate professor who’s also been on the Internet forever. Both the owner’s academic background and the fact that these domains are directly registered by him and not through a third-party registrar makes me think these domains might be planned for some sort of Internet infrastructure. However, it’s anyone’s guess as to exactly what these domains will be used for (if anything at all).

14

This appears to be a Tumblr blog stylized like a retro Windows 98 desktop. The blog itself appears somewhat young and devoid of content. Most of the posts seem to be community-submitted and are either AI art showcases or ads. There’s not much else to say about this one other than the fact that it’s 1000x more interesting than either of its neighboring domains.

255

This site returns a JSON response, claiming to be a site for robots. However, further entry to the site requires solving an “anti-human CAPTCHA” first. The challenge in question is “stopwar” which I found to be both hilarious and depressing at the same time. I do wonder what would happen to the site if humanity collectively stopped all conflicts (or the more likely scenario, when humanity eventually does die out).

418

Claims to be HTCPCP compliant server. HTCPCP is a facetious protocol defined in a 1998 April Fool’s RFC. The protocol is infamous for having introduced Status 418: I’m a teapot which is of course also the very number in the domain name. Whether or not this server actually brews real coffee remains a mystery.

455

This appears to be an AI app of some kind (the marketing homepage is somewhat vague about exactly what, and I’m not signing up for an account to find out). What really caught my attention for this one was the EVA themed 404 page though. Pretty. Cool.

727

This is a redirect to a Vtuber’s Twitch. The Vtuber in question is called Neuro-sama. Unlike most Vtubers, however, Neuro is not a human; it’s an AI. The Vtuber was created by a guy named Vedal and functionally appears to be an LLM strapped to text-to-speech with a few more add-ons and AI systems attached that allow her to do things like respond to her creator and Twitch chat as well as play games like osu!.

What surprised me the most about Neuro is how popular she is. Apparently, she is the 7th most subscribed channel on Twitch as of January 2025. For all the criticism that AI generated entertainment gets, I found it odd to see an AI Vtuber on the other end of this spectrum, getting lots of love and attention from the public. At first I chalked it up to just the passing interest of others towards the novelty of having an LLM as a Vtuber. However, after watching a few of Neuro’s stream compilations as research for this blog post, I actually found myself enjoying her content quite a lot. Even though Neuro’s voice and personality give off a very robotic and artificial vibe, there’s some sort of inexplicable charm to her absurdity. I found myself laughing along as I listened to the banter between her, Evil Neuro, and their creator, Vedal. Personally, for all the hype in AI products today, I feel that it is perhaps AI companions like Neuro which will end up surviving in the long run.

After all this, one major question still remained for me though, and that was what in the world does Neuro have to do with the number 727? Perhaps it’s just some inside joke I don’t get? I guess I’d have to become her fan to find out.

Closing words

While this journey through cyberspace from 1 to 1000 was mostly just a bunch of barren pages and ads, there were enough small gems to make it worth my while.

It was particularly interesting to see the wide array of applications people were making use of their .ai domains for, whether that be selling real products, making jokes, promoting ideas and idols, or the tried-and-true practice of shilling crypto, each of them felt like a window into someone else’s world. I think it just goes to show that even in 2025, a time where people lament the collapse of the old Internet and the centralization of media on corporate-owned platforms like Instagram or Twitter/X, there are still fun little worlds floating out there in the Ether waiting to be discovered.

Anyways, until next time!

Appendix A: On Visiting Random Websites

Trawling down a list of unknown websites is kinda like trying a series of doors on a row of buildings. Sometimes you open the door, and it’s a quirky little coffee shop while other times, it’s a dank and dusty crack den. All this is to say that visiting a slew of random websites in Selenium probably wasn’t the smartest thing I’ve done so far this year.

Many of the websites I encountered were luckily just parking pages, but there were definitely a number of extremely shady sites. Even though I knew from a technical standpoint that the Selenium browser session is separate from my personal one, so any attempt to steal cookies or run CSRF probably didn’t work, the whole thing still just made me want to scrub my laptop down.

Appendix B: The Actual Answer

As I mentioned before, I found this puzzle off of Lobsters. I imagine so did a lot of other people. Typically, throwing a bunch of brains at a common problem often has the uncanny result of solving it. I thought I might as well document how the rest of the puzzle solving played out (since I didn’t really contribute to solving it at all).

Most of this puzzle was actually already figured out by the author of the original blog post. However, a few details were missing or off:

• BusyBeaver(4). The answer they wanted was the one that gives 13.
• log*(16). Apparently this means the iterated log2 of 16 which turns out to be 3.
• The prefix function. This seemed to have been the major stumper in the original blog post. Another Lobster had thought it meant the Knuth-Morris-Pratt prefix function, but that didn’t work either. Eventually someone(s) figured out that it actually meant the SI prefix of the input.

Putting it all together we get:

prefix(
    [
        6
        + argmax(
            softmax(
                [0.693147181 , 5.85987448]
            )
        )
        + 3
    ] ^ [
        13
        + 4
        + 1
    ]
)

.ai/givemeprize

Which reduces to:

prefix(10^18)

.ai/givemeprize

The SI prefix for this is exa.

Visiting exa.ai/givemeprize results in a job ad. In all fairness, this is not an unexpected result considering that the problem originated from a flyer on a telephone pole.

In other news, it looks like ChatGPT managed to solve this puzzle as well. Maybe someone should hire it :P.